]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/kern/task_policy.c
xnu-7195.101.1.tar.gz
[apple/xnu.git] / osfmk / kern / task_policy.c
index c382f56ace0bf0b61f14e04fe230a096827ee2c0..b9664280336530631e13311409985d394a1134b0 100644 (file)
@@ -1,17 +1,20 @@
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2020 Apple Computer, Inc. All rights reserved.
+ *
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  *
- * @APPLE_LICENSE_HEADER_START@
- * 
- * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
- * 
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
- * compliance with the License. Please obtain a copy of the License at
- * http://www.opensource.apple.com/apsl/ and read it before using this
- * file.
- * 
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ *
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ *
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  * Please see the License for the specific language governing rights and
  * limitations under the License.
- * 
- * @APPLE_LICENSE_HEADER_END@
+ *
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
+
+#include <kern/policy_internal.h>
+#include <mach/task_policy.h>
+
+#include <mach/mach_types.h>
+#include <mach/task_server.h>
+
+#include <kern/host.h>                  /* host_priv_self()        */
+#include <mach/host_priv.h>             /* host_get_special_port() */
+#include <mach/host_special_ports.h>    /* RESOURCE_NOTIFY_PORT    */
+#include <kern/sched.h>
+#include <kern/task.h>
+#include <mach/thread_policy.h>
+#include <sys/errno.h>
+#include <sys/resource.h>
+#include <machine/limits.h>
+#include <kern/ledger.h>
+#include <kern/thread_call.h>
+#include <kern/sfi.h>
+#include <kern/coalition.h>
+#if CONFIG_TELEMETRY
+#include <kern/telemetry.h>
+#endif
+#if !defined(XNU_TARGET_OS_OSX)
+#include <kern/kalloc.h>
+#include <sys/errno.h>
+#endif /* !defined(XNU_TARGET_OS_OSX) */
+
+#if IMPORTANCE_INHERITANCE
+#include <ipc/ipc_importance.h>
+#if IMPORTANCE_TRACE
+#include <mach/machine/sdt.h>
+#endif /* IMPORTANCE_TRACE */
+#endif /* IMPORTANCE_INHERITACE */
+
+#include <sys/kdebug.h>
+
 /*
- * Copyright (c) 2000 Apple Computer, Inc.  All rights reserved.
+ *  Task Policy
+ *
+ *  This subsystem manages task and thread IO priority and backgrounding,
+ *  as well as importance inheritance, process suppression, task QoS, and apptype.
+ *  These properties have a suprising number of complex interactions, so they are
+ *  centralized here in one state machine to simplify the implementation of those interactions.
+ *
+ *  Architecture:
+ *  Threads and tasks have two policy fields: requested, effective.
+ *  Requested represents the wishes of each interface that influences task policy.
+ *  Effective represents the distillation of that policy into a set of behaviors.
+ *
+ *  Each thread making a modification in the policy system passes a 'pending' struct,
+ *  which tracks updates that will be applied after dropping the policy engine lock.
+ *
+ *  Each interface that has an input into the task policy state machine controls a field in requested.
+ *  If the interface has a getter, it returns what is in the field in requested, but that is
+ *  not necessarily what is actually in effect.
+ *
+ *  All kernel subsystems that behave differently based on task policy call into
+ *  the proc_get_effective_(task|thread)_policy functions, which return the decision of the task policy state machine
+ *  for that subsystem by querying only the 'effective' field.
  *
- * HISTORY
+ *  Policy change operations:
+ *  Here are the steps to change a policy on a task or thread:
+ *  1) Lock task
+ *  2) Change requested field for the relevant policy
+ *  3) Run a task policy update, which recalculates effective based on requested,
+ *     then takes a diff between the old and new versions of requested and calls the relevant
+ *     other subsystems to apply these changes, and updates the pending field.
+ *  4) Unlock task
+ *  5) Run task policy update complete, which looks at the pending field to update
+ *     subsystems which cannot be touched while holding the task lock.
  *
- * 15 October 2000 (debo)
- *  Created.
+ *  To add a new requested policy, add the field in the requested struct, the flavor in task.h,
+ *  the setter and getter in proc_(set|get)_task_policy*,
+ *  then set up the effects of that behavior in task_policy_update*. If the policy manifests
+ *  itself as a distinct effective policy, add it to the effective struct and add it to the
+ *  proc_get_effective_task_policy accessor.
+ *
+ *  Most policies are set via proc_set_task_policy, but policies that don't fit that interface
+ *  roll their own lock/set/update/unlock/complete code inside this file.
+ *
+ *
+ *  Suppression policy
+ *
+ *  These are a set of behaviors that can be requested for a task.  They currently have specific
+ *  implied actions when they're enabled, but they may be made customizable in the future.
+ *
+ *  When the affected task is boosted, we temporarily disable the suppression behaviors
+ *  so that the affected process has a chance to run so it can call the API to permanently
+ *  disable the suppression behaviors.
+ *
+ *  Locking
+ *
+ *  Changing task policy on a task takes the task lock.
+ *  Changing task policy on a thread takes the thread mutex.
+ *  Task policy changes that affect threads will take each thread's mutex to update it if necessary.
+ *
+ *  Querying the effective policy does not take a lock, because callers
+ *  may run in interrupt context or other place where locks are not OK.
+ *
+ *  This means that any notification of state change needs to be externally synchronized.
+ *  We do this by idempotent callouts after the state has changed to ask
+ *  other subsystems to update their view of the world.
+ *
+ * TODO: Move all cpu/wakes/io monitor code into a separate file
+ * TODO: Move all importance code over to importance subsystem
+ * TODO: Move all taskwatch code into a separate file
+ * TODO: Move all VM importance code into a separate file
  */
 
-#include <kern/task.h>
+/* Task policy related helper functions */
+static void proc_set_task_policy_locked(task_t task, int category, int flavor, int value, int value2);
+
+static void task_policy_update_locked(task_t task, task_pend_token_t pend_token);
+static void task_policy_update_internal_locked(task_t task, bool in_create, task_pend_token_t pend_token);
+
+/* For attributes that have two scalars as input/output */
+static void proc_set_task_policy2(task_t task, int category, int flavor, int value1, int value2);
+static void proc_get_task_policy2(task_t task, int category, int flavor, int *value1, int *value2);
+
+static boolean_t task_policy_update_coalition_focal_tasks(task_t task, int prev_role, int next_role, task_pend_token_t pend_token);
+
+static uint64_t task_requested_bitfield(task_t task);
+static uint64_t task_effective_bitfield(task_t task);
+
+/* Convenience functions for munging a policy bitfield into a tracepoint */
+static uintptr_t trequested_0(task_t task);
+static uintptr_t trequested_1(task_t task);
+static uintptr_t teffective_0(task_t task);
+static uintptr_t teffective_1(task_t task);
+
+/* CPU limits helper functions */
+static int task_set_cpuusage(task_t task, uint8_t percentage, uint64_t interval, uint64_t deadline, int scope, int entitled);
+static int task_get_cpuusage(task_t task, uint8_t *percentagep, uint64_t *intervalp, uint64_t *deadlinep, int *scope);
+static int task_enable_cpumon_locked(task_t task);
+static int task_disable_cpumon(task_t task);
+static int task_clear_cpuusage_locked(task_t task, int cpumon_entitled);
+static int task_apply_resource_actions(task_t task, int type);
+static void task_action_cpuusage(thread_call_param_t param0, thread_call_param_t param1);
+
+#ifdef MACH_BSD
+typedef struct proc *   proc_t;
+int                     proc_pid(struct proc *proc);
+extern int              proc_selfpid(void);
+extern char *           proc_name_address(void *p);
+extern char *           proc_best_name(proc_t proc);
+
+extern int proc_pidpathinfo_internal(proc_t p, uint64_t arg,
+    char *buffer, uint32_t buffersize,
+    int32_t *retval);
+#endif /* MACH_BSD */
+
+
+#if CONFIG_TASKWATCH
+/* Taskwatch related helper functions */
+static void set_thread_appbg(thread_t thread, int setbg, int importance);
+static void add_taskwatch_locked(task_t task, task_watch_t * twp);
+static void remove_taskwatch_locked(task_t task, task_watch_t * twp);
+static void task_watch_lock(void);
+static void task_watch_unlock(void);
+static void apply_appstate_watchers(task_t task);
+
+typedef struct task_watcher {
+       queue_chain_t   tw_links;       /* queueing of threads */
+       task_t          tw_task;        /* task that is being watched */
+       thread_t        tw_thread;      /* thread that is watching the watch_task */
+       int             tw_state;       /* the current app state of the thread */
+       int             tw_importance;  /* importance prior to backgrounding */
+} task_watch_t;
+
+typedef struct thread_watchlist {
+       thread_t        thread;         /* thread being worked on for taskwatch action */
+       int             importance;     /* importance to be restored if thread is being made active */
+} thread_watchlist_t;
+
+#endif /* CONFIG_TASKWATCH */
+
+extern int memorystatus_update_priority_for_appnap(proc_t p, boolean_t is_appnap);
+
+/* Importance Inheritance related helper functions */
+
+#if IMPORTANCE_INHERITANCE
+
+static void task_importance_mark_live_donor(task_t task, boolean_t donating);
+static void task_importance_mark_receiver(task_t task, boolean_t receiving);
+static void task_importance_mark_denap_receiver(task_t task, boolean_t denap);
+
+static boolean_t task_is_marked_live_importance_donor(task_t task);
+static boolean_t task_is_importance_receiver(task_t task);
+static boolean_t task_is_importance_denap_receiver(task_t task);
+
+static int task_importance_hold_internal_assertion(task_t target_task, uint32_t count);
+
+static void task_add_importance_watchport(task_t task, mach_port_t port, int *boostp);
+static void task_importance_update_live_donor(task_t target_task);
+
+static void task_set_boost_locked(task_t task, boolean_t boost_active);
+
+#endif /* IMPORTANCE_INHERITANCE */
+
+#if IMPORTANCE_TRACE
+#define __imptrace_only
+#else /* IMPORTANCE_TRACE */
+#define __imptrace_only __unused
+#endif /* !IMPORTANCE_TRACE */
+
+#if IMPORTANCE_INHERITANCE
+#define __imp_only
+#else
+#define __imp_only __unused
+#endif
+
+/*
+ * Default parameters for certain policies
+ */
+
+int proc_standard_daemon_tier = THROTTLE_LEVEL_TIER1;
+int proc_suppressed_disk_tier = THROTTLE_LEVEL_TIER1;
+int proc_tal_disk_tier        = THROTTLE_LEVEL_TIER1;
+
+int proc_graphics_timer_qos   = (LATENCY_QOS_TIER_0 & 0xFF);
+
+const int proc_default_bg_iotier  = THROTTLE_LEVEL_TIER2;
+
+/* Latency/throughput QoS fields remain zeroed, i.e. TIER_UNSPECIFIED at creation */
+const struct task_requested_policy default_task_requested_policy = {
+       .trp_bg_iotier = proc_default_bg_iotier
+};
+const struct task_effective_policy default_task_effective_policy = {};
+
+/*
+ * Default parameters for CPU usage monitor.
+ *
+ * Default setting is 50% over 3 minutes.
+ */
+#define         DEFAULT_CPUMON_PERCENTAGE 50
+#define         DEFAULT_CPUMON_INTERVAL   (3 * 60)
+
+uint8_t         proc_max_cpumon_percentage;
+uint64_t        proc_max_cpumon_interval;
 
-static void
-task_priority(
-       task_t                  task,
-       integer_t               priority,
-       integer_t               max_priority);
+
+kern_return_t
+qos_latency_policy_validate(task_latency_qos_t ltier)
+{
+       if ((ltier != LATENCY_QOS_TIER_UNSPECIFIED) &&
+           ((ltier > LATENCY_QOS_TIER_5) || (ltier < LATENCY_QOS_TIER_0))) {
+               return KERN_INVALID_ARGUMENT;
+       }
+
+       return KERN_SUCCESS;
+}
+
+kern_return_t
+qos_throughput_policy_validate(task_throughput_qos_t ttier)
+{
+       if ((ttier != THROUGHPUT_QOS_TIER_UNSPECIFIED) &&
+           ((ttier > THROUGHPUT_QOS_TIER_5) || (ttier < THROUGHPUT_QOS_TIER_0))) {
+               return KERN_INVALID_ARGUMENT;
+       }
+
+       return KERN_SUCCESS;
+}
+
+static kern_return_t
+task_qos_policy_validate(task_qos_policy_t qosinfo, mach_msg_type_number_t count)
+{
+       if (count < TASK_QOS_POLICY_COUNT) {
+               return KERN_INVALID_ARGUMENT;
+       }
+
+       task_latency_qos_t ltier = qosinfo->task_latency_qos_tier;
+       task_throughput_qos_t ttier = qosinfo->task_throughput_qos_tier;
+
+       kern_return_t kr = qos_latency_policy_validate(ltier);
+
+       if (kr != KERN_SUCCESS) {
+               return kr;
+       }
+
+       kr = qos_throughput_policy_validate(ttier);
+
+       return kr;
+}
+
+uint32_t
+qos_extract(uint32_t qv)
+{
+       return qv & 0xFF;
+}
+
+uint32_t
+qos_latency_policy_package(uint32_t qv)
+{
+       return (qv == LATENCY_QOS_TIER_UNSPECIFIED) ? LATENCY_QOS_TIER_UNSPECIFIED : ((0xFF << 16) | qv);
+}
+
+uint32_t
+qos_throughput_policy_package(uint32_t qv)
+{
+       return (qv == THROUGHPUT_QOS_TIER_UNSPECIFIED) ? THROUGHPUT_QOS_TIER_UNSPECIFIED : ((0xFE << 16) | qv);
+}
+
+#define TASK_POLICY_SUPPRESSION_DISABLE  0x1
+#define TASK_POLICY_SUPPRESSION_IOTIER2  0x2
+#define TASK_POLICY_SUPPRESSION_NONDONOR 0x4
+/* TEMPORARY boot-arg controlling task_policy suppression (App Nap) */
+static boolean_t task_policy_suppression_flags = TASK_POLICY_SUPPRESSION_IOTIER2 |
+    TASK_POLICY_SUPPRESSION_NONDONOR;
 
 kern_return_t
 task_policy_set(
-       task_t                                  task,
-       task_policy_flavor_t    flavor,
-       task_policy_t                   policy_info,
-       mach_msg_type_number_t  count)
+       task_t                                  task,
+       task_policy_flavor_t    flavor,
+       task_policy_t                   policy_info,
+       mach_msg_type_number_t  count)
 {
-       kern_return_t           result = KERN_SUCCESS;
+       kern_return_t           result = KERN_SUCCESS;
 
-       if (task == TASK_NULL || task == kernel_task)
-               return (KERN_INVALID_ARGUMENT);
+       if (task == TASK_NULL || task == kernel_task) {
+               return KERN_INVALID_ARGUMENT;
+       }
 
        switch (flavor) {
+       case TASK_CATEGORY_POLICY: {
+               task_category_policy_t info = (task_category_policy_t)policy_info;
 
-       case TASK_CATEGORY_POLICY:
-       {
-               task_category_policy_t          info = (task_category_policy_t)policy_info;
+               if (count < TASK_CATEGORY_POLICY_COUNT) {
+                       return KERN_INVALID_ARGUMENT;
+               }
 
-               if (count < TASK_CATEGORY_POLICY_COUNT)
-                       return (KERN_INVALID_ARGUMENT);
+#if !defined(XNU_TARGET_OS_OSX)
+               /* On embedded, you can't modify your own role. */
+               if (current_task() == task) {
+                       return KERN_INVALID_ARGUMENT;
+               }
+#endif
 
-               task_lock(task);
+               switch (info->role) {
+               case TASK_FOREGROUND_APPLICATION:
+               case TASK_BACKGROUND_APPLICATION:
+               case TASK_DEFAULT_APPLICATION:
+                       proc_set_task_policy(task,
+                           TASK_POLICY_ATTRIBUTE, TASK_POLICY_ROLE,
+                           info->role);
+                       break;
 
-               if (    info->role == TASK_FOREGROUND_APPLICATION       ||
-                               info->role == TASK_BACKGROUND_APPLICATION               ) {
-                       switch (task->role) {
-
-                       case TASK_FOREGROUND_APPLICATION:
-                       case TASK_BACKGROUND_APPLICATION:
-                       case TASK_UNSPECIFIED:
-                               task_priority(task,
-                                                               ((info->role == TASK_FOREGROUND_APPLICATION)?
-                                                                       BASEPRI_FOREGROUND: BASEPRI_BACKGROUND),
-                                                         task->max_priority);
-                               task->role = info->role;
-                               break;
-
-                       case TASK_CONTROL_APPLICATION:
-                       case TASK_RENICED:
-                               /* fail silently */
-                               break;
-
-                       default:
+               case TASK_CONTROL_APPLICATION:
+                       if (task != current_task() || task->sec_token.val[0] != 0) {
                                result = KERN_INVALID_ARGUMENT;
-                               break;
+                       } else {
+                               proc_set_task_policy(task,
+                                   TASK_POLICY_ATTRIBUTE, TASK_POLICY_ROLE,
+                                   info->role);
                        }
-               }
-               else
-               if (info->role == TASK_CONTROL_APPLICATION) {
-                       if (    task != current_task()                  ||
-                                       task->sec_token.val[0] != 0                     )
+                       break;
+
+               case TASK_GRAPHICS_SERVER:
+                       /* TODO: Restrict this role to FCFS <rdar://problem/12552788> */
+                       if (task != current_task() || task->sec_token.val[0] != 0) {
                                result = KERN_INVALID_ARGUMENT;
-                       else {
-                               task_priority(task, BASEPRI_CONTROL, task->max_priority);
-                               task->role = info->role;
+                       } else {
+                               proc_set_task_policy(task,
+                                   TASK_POLICY_ATTRIBUTE, TASK_POLICY_ROLE,
+                                   info->role);
                        }
+                       break;
+               default:
+                       result = KERN_INVALID_ARGUMENT;
+                       break;
+               } /* switch (info->role) */
+
+               break;
+       }
+
+/* Desired energy-efficiency/performance "quality-of-service" */
+       case TASK_BASE_QOS_POLICY:
+       case TASK_OVERRIDE_QOS_POLICY:
+       {
+               task_qos_policy_t qosinfo = (task_qos_policy_t)policy_info;
+               kern_return_t kr = task_qos_policy_validate(qosinfo, count);
+
+               if (kr != KERN_SUCCESS) {
+                       return kr;
                }
-               else
-               if (info->role == TASK_GRAPHICS_SERVER) {
-                       if (    task != current_task()                  ||
-                                       task->sec_token.val[0] != 0                     )
-                               result = KERN_INVALID_ARGUMENT;
-                       else {
-                               task_priority(task, MAXPRI_SYSTEM - 3, MAXPRI_SYSTEM);
-                               task->role = info->role;
-                       }
+
+
+               uint32_t lqos = qos_extract(qosinfo->task_latency_qos_tier);
+               uint32_t tqos = qos_extract(qosinfo->task_throughput_qos_tier);
+
+               proc_set_task_policy2(task, TASK_POLICY_ATTRIBUTE,
+                   flavor == TASK_BASE_QOS_POLICY ? TASK_POLICY_BASE_LATENCY_AND_THROUGHPUT_QOS : TASK_POLICY_OVERRIDE_LATENCY_AND_THROUGHPUT_QOS,
+                   lqos, tqos);
+       }
+       break;
+
+       case TASK_BASE_LATENCY_QOS_POLICY:
+       {
+               task_qos_policy_t qosinfo = (task_qos_policy_t)policy_info;
+               kern_return_t kr = task_qos_policy_validate(qosinfo, count);
+
+               if (kr != KERN_SUCCESS) {
+                       return kr;
                }
-               else
-                       result = KERN_INVALID_ARGUMENT;
 
-               task_unlock(task);
+               uint32_t lqos = qos_extract(qosinfo->task_latency_qos_tier);
 
-               break;
+               proc_set_task_policy(task, TASK_POLICY_ATTRIBUTE, TASK_BASE_LATENCY_QOS_POLICY, lqos);
        }
+       break;
 
-       default:
+       case TASK_BASE_THROUGHPUT_QOS_POLICY:
+       {
+               task_qos_policy_t qosinfo = (task_qos_policy_t)policy_info;
+               kern_return_t kr = task_qos_policy_validate(qosinfo, count);
+
+               if (kr != KERN_SUCCESS) {
+                       return kr;
+               }
+
+               uint32_t tqos = qos_extract(qosinfo->task_throughput_qos_tier);
+
+               proc_set_task_policy(task, TASK_POLICY_ATTRIBUTE, TASK_BASE_THROUGHPUT_QOS_POLICY, tqos);
+       }
+       break;
+
+       case TASK_SUPPRESSION_POLICY:
+       {
+#if !defined(XNU_TARGET_OS_OSX)
+               /*
+                * Suppression policy is not enabled for embedded
+                * because apps aren't marked as denap receivers
+                */
                result = KERN_INVALID_ARGUMENT;
                break;
-       }
+#else /* !defined(XNU_TARGET_OS_OSX) */
 
-       return (result);
-}
+               task_suppression_policy_t info = (task_suppression_policy_t)policy_info;
 
-static void
-task_priority(
-       task_t                  task,
-       integer_t               priority,
-       integer_t               max_priority)
-{
-       thread_act_t    act;
+               if (count < TASK_SUPPRESSION_POLICY_COUNT) {
+                       return KERN_INVALID_ARGUMENT;
+               }
+
+               struct task_qos_policy qosinfo;
+
+               qosinfo.task_latency_qos_tier = info->timer_throttle;
+               qosinfo.task_throughput_qos_tier = info->throughput_qos;
+
+               kern_return_t kr = task_qos_policy_validate(&qosinfo, TASK_QOS_POLICY_COUNT);
+
+               if (kr != KERN_SUCCESS) {
+                       return kr;
+               }
+
+               /* TEMPORARY disablement of task suppression */
+               if (info->active &&
+                   (task_policy_suppression_flags & TASK_POLICY_SUPPRESSION_DISABLE)) {
+                       return KERN_SUCCESS;
+               }
 
-       task->max_priority = max_priority;
+               struct task_pend_token pend_token = {};
 
-       if (priority > task->max_priority)
-               priority = task->max_priority;
-       else
-       if (priority < MINPRI)
-               priority = MINPRI;
+               task_lock(task);
+
+               KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                   (IMPORTANCE_CODE(IMP_TASK_SUPPRESSION, info->active)) | DBG_FUNC_START,
+                   proc_selfpid(), task_pid(task), trequested_0(task),
+                   trequested_1(task), 0);
+
+               task->requested_policy.trp_sup_active      = (info->active)         ? 1 : 0;
+               task->requested_policy.trp_sup_lowpri_cpu  = (info->lowpri_cpu)     ? 1 : 0;
+               task->requested_policy.trp_sup_timer       = qos_extract(info->timer_throttle);
+               task->requested_policy.trp_sup_disk        = (info->disk_throttle)  ? 1 : 0;
+               task->requested_policy.trp_sup_throughput  = qos_extract(info->throughput_qos);
+               task->requested_policy.trp_sup_cpu         = (info->suppressed_cpu) ? 1 : 0;
+               task->requested_policy.trp_sup_bg_sockets  = (info->background_sockets) ? 1 : 0;
+
+               task_policy_update_locked(task, &pend_token);
 
-       task->priority = priority;
+               KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                   (IMPORTANCE_CODE(IMP_TASK_SUPPRESSION, info->active)) | DBG_FUNC_END,
+                   proc_selfpid(), task_pid(task), trequested_0(task),
+                   trequested_1(task), 0);
 
-       queue_iterate(&task->threads, act, thread_act_t, task_threads) {
-               thread_t                thread = act_lock_thread(act);
+               task_unlock(task);
+
+               task_policy_update_complete_unlocked(task, &pend_token);
+
+               break;
 
-               if (act->active)
-                       thread_task_priority(thread, priority, max_priority);
+#endif /* !defined(XNU_TARGET_OS_OSX) */
+       }
 
-               act_unlock_thread(act);
+       default:
+               result = KERN_INVALID_ARGUMENT;
+               break;
        }
+
+       return result;
 }
 
+/* Sets BSD 'nice' value on the task */
 kern_return_t
 task_importance(
-       task_t                          task,
-       integer_t                       importance)
+       task_t                          task,
+       integer_t                       importance)
 {
-       if (task == TASK_NULL || task == kernel_task)
-               return (KERN_INVALID_ARGUMENT);
+       if (task == TASK_NULL || task == kernel_task) {
+               return KERN_INVALID_ARGUMENT;
+       }
 
        task_lock(task);
 
        if (!task->active) {
                task_unlock(task);
 
-               return (KERN_TERMINATED);
+               return KERN_TERMINATED;
        }
 
-       if (task->role >= TASK_CONTROL_APPLICATION) {
+       if (proc_get_effective_task_policy(task, TASK_POLICY_ROLE) >= TASK_CONTROL_APPLICATION) {
                task_unlock(task);
 
-               return (KERN_INVALID_ARGUMENT);
+               return KERN_INVALID_ARGUMENT;
        }
 
-       task_priority(task, importance + BASEPRI_DEFAULT, task->max_priority);
-       task->role = TASK_RENICED;
+       task->importance = importance;
+
+       struct task_pend_token pend_token = {};
+
+       task_policy_update_locked(task, &pend_token);
 
        task_unlock(task);
 
-       return (KERN_SUCCESS);
+       task_policy_update_complete_unlocked(task, &pend_token);
+
+       return KERN_SUCCESS;
 }
-               
+
 kern_return_t
 task_policy_get(
-       task_t                                  task,
-       task_policy_flavor_t    flavor,
-       task_policy_t                   policy_info,
-       mach_msg_type_number_t  *count,
-       boolean_t                               *get_default)
+       task_t                                  task,
+       task_policy_flavor_t    flavor,
+       task_policy_t                   policy_info,
+       mach_msg_type_number_t  *count,
+       boolean_t                               *get_default)
 {
-       if (task == TASK_NULL || task == kernel_task)
-               return (KERN_INVALID_ARGUMENT);
+       if (task == TASK_NULL || task == kernel_task) {
+               return KERN_INVALID_ARGUMENT;
+       }
 
        switch (flavor) {
-
        case TASK_CATEGORY_POLICY:
        {
-               task_category_policy_t          info = (task_category_policy_t)policy_info;
+               task_category_policy_t          info = (task_category_policy_t)policy_info;
 
-               if (*count < TASK_CATEGORY_POLICY_COUNT)
-                       return (KERN_INVALID_ARGUMENT);
+               if (*count < TASK_CATEGORY_POLICY_COUNT) {
+                       return KERN_INVALID_ARGUMENT;
+               }
 
-               if (*get_default)
+               if (*get_default) {
                        info->role = TASK_UNSPECIFIED;
-               else {
+               } else {
+                       info->role = proc_get_task_policy(task, TASK_POLICY_ATTRIBUTE, TASK_POLICY_ROLE);
+               }
+               break;
+       }
+
+       case TASK_BASE_QOS_POLICY: /* FALLTHRU */
+       case TASK_OVERRIDE_QOS_POLICY:
+       {
+               task_qos_policy_t info = (task_qos_policy_t)policy_info;
+
+               if (*count < TASK_QOS_POLICY_COUNT) {
+                       return KERN_INVALID_ARGUMENT;
+               }
+
+               if (*get_default) {
+                       info->task_latency_qos_tier = LATENCY_QOS_TIER_UNSPECIFIED;
+                       info->task_throughput_qos_tier = THROUGHPUT_QOS_TIER_UNSPECIFIED;
+               } else if (flavor == TASK_BASE_QOS_POLICY) {
+                       int value1, value2;
+
+                       proc_get_task_policy2(task, TASK_POLICY_ATTRIBUTE, TASK_POLICY_BASE_LATENCY_AND_THROUGHPUT_QOS, &value1, &value2);
+
+                       info->task_latency_qos_tier = qos_latency_policy_package(value1);
+                       info->task_throughput_qos_tier = qos_throughput_policy_package(value2);
+               } else if (flavor == TASK_OVERRIDE_QOS_POLICY) {
+                       int value1, value2;
+
+                       proc_get_task_policy2(task, TASK_POLICY_ATTRIBUTE, TASK_POLICY_OVERRIDE_LATENCY_AND_THROUGHPUT_QOS, &value1, &value2);
+
+                       info->task_latency_qos_tier = qos_latency_policy_package(value1);
+                       info->task_throughput_qos_tier = qos_throughput_policy_package(value2);
+               }
+
+               break;
+       }
+
+       case TASK_POLICY_STATE:
+       {
+               task_policy_state_t info = (task_policy_state_t)policy_info;
+
+               if (*count < TASK_POLICY_STATE_COUNT) {
+                       return KERN_INVALID_ARGUMENT;
+               }
+
+               /* Only root can get this info */
+               if (current_task()->sec_token.val[0] != 0) {
+                       return KERN_PROTECTION_FAILURE;
+               }
+
+               if (*get_default) {
+                       info->requested = 0;
+                       info->effective = 0;
+                       info->pending = 0;
+                       info->imp_assertcnt = 0;
+                       info->imp_externcnt = 0;
+                       info->flags = 0;
+                       info->imp_transitions = 0;
+               } else {
                        task_lock(task);
-                       info->role = task->role;
+
+                       info->requested = task_requested_bitfield(task);
+                       info->effective = task_effective_bitfield(task);
+                       info->pending   = 0;
+
+                       info->tps_requested_policy = *(uint64_t*)(&task->requested_policy);
+                       info->tps_effective_policy = *(uint64_t*)(&task->effective_policy);
+
+                       info->flags = 0;
+                       if (task->task_imp_base != NULL) {
+                               info->imp_assertcnt = task->task_imp_base->iit_assertcnt;
+                               info->imp_externcnt = IIT_EXTERN(task->task_imp_base);
+                               info->flags |= (task_is_marked_importance_receiver(task) ? TASK_IMP_RECEIVER : 0);
+                               info->flags |= (task_is_marked_importance_denap_receiver(task) ? TASK_DENAP_RECEIVER : 0);
+                               info->flags |= (task_is_marked_importance_donor(task) ? TASK_IMP_DONOR : 0);
+                               info->flags |= (task_is_marked_live_importance_donor(task) ? TASK_IMP_LIVE_DONOR : 0);
+                               info->flags |= (get_task_pidsuspended(task) ? TASK_IS_PIDSUSPENDED : 0);
+                               info->imp_transitions = task->task_imp_base->iit_transitions;
+                       } else {
+                               info->imp_assertcnt = 0;
+                               info->imp_externcnt = 0;
+                               info->imp_transitions = 0;
+                       }
                        task_unlock(task);
                }
+
+               break;
+       }
+
+       case TASK_SUPPRESSION_POLICY:
+       {
+               task_suppression_policy_t info = (task_suppression_policy_t)policy_info;
+
+               if (*count < TASK_SUPPRESSION_POLICY_COUNT) {
+                       return KERN_INVALID_ARGUMENT;
+               }
+
+               task_lock(task);
+
+               if (*get_default) {
+                       info->active            = 0;
+                       info->lowpri_cpu        = 0;
+                       info->timer_throttle    = LATENCY_QOS_TIER_UNSPECIFIED;
+                       info->disk_throttle     = 0;
+                       info->cpu_limit         = 0;
+                       info->suspend           = 0;
+                       info->throughput_qos    = 0;
+                       info->suppressed_cpu    = 0;
+               } else {
+                       info->active            = task->requested_policy.trp_sup_active;
+                       info->lowpri_cpu        = task->requested_policy.trp_sup_lowpri_cpu;
+                       info->timer_throttle    = qos_latency_policy_package(task->requested_policy.trp_sup_timer);
+                       info->disk_throttle     = task->requested_policy.trp_sup_disk;
+                       info->cpu_limit         = 0;
+                       info->suspend           = 0;
+                       info->throughput_qos    = qos_throughput_policy_package(task->requested_policy.trp_sup_throughput);
+                       info->suppressed_cpu    = task->requested_policy.trp_sup_cpu;
+                       info->background_sockets = task->requested_policy.trp_sup_bg_sockets;
+               }
+
+               task_unlock(task);
                break;
        }
 
        default:
-               return (KERN_INVALID_ARGUMENT);
+               return KERN_INVALID_ARGUMENT;
+       }
+
+       return KERN_SUCCESS;
+}
+
+/*
+ * Called at task creation
+ * We calculate the correct effective but don't apply it to anything yet.
+ * The threads, etc will inherit from the task as they get created.
+ */
+void
+task_policy_create(task_t task, task_t parent_task)
+{
+       task->requested_policy.trp_apptype          = parent_task->requested_policy.trp_apptype;
+
+       task->requested_policy.trp_int_darwinbg     = parent_task->requested_policy.trp_int_darwinbg;
+       task->requested_policy.trp_ext_darwinbg     = parent_task->requested_policy.trp_ext_darwinbg;
+       task->requested_policy.trp_int_iotier       = parent_task->requested_policy.trp_int_iotier;
+       task->requested_policy.trp_ext_iotier       = parent_task->requested_policy.trp_ext_iotier;
+       task->requested_policy.trp_int_iopassive    = parent_task->requested_policy.trp_int_iopassive;
+       task->requested_policy.trp_ext_iopassive    = parent_task->requested_policy.trp_ext_iopassive;
+       task->requested_policy.trp_bg_iotier        = parent_task->requested_policy.trp_bg_iotier;
+       task->requested_policy.trp_terminated       = parent_task->requested_policy.trp_terminated;
+       task->requested_policy.trp_qos_clamp        = parent_task->requested_policy.trp_qos_clamp;
+
+       if (task->requested_policy.trp_apptype == TASK_APPTYPE_DAEMON_ADAPTIVE && !task_is_exec_copy(task)) {
+               /* Do not update the apptype for exec copy task */
+               if (parent_task->requested_policy.trp_boosted) {
+                       task->requested_policy.trp_apptype = TASK_APPTYPE_DAEMON_INTERACTIVE;
+                       task_importance_mark_donor(task, TRUE);
+               } else {
+                       task->requested_policy.trp_apptype = TASK_APPTYPE_DAEMON_BACKGROUND;
+                       task_importance_mark_receiver(task, FALSE);
+               }
+       }
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           (IMPORTANCE_CODE(IMP_UPDATE, (IMP_UPDATE_TASK_CREATE | TASK_POLICY_TASK))) | DBG_FUNC_START,
+           task_pid(task), teffective_0(task),
+           teffective_1(task), task->priority, 0);
+
+       task_policy_update_internal_locked(task, true, NULL);
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           (IMPORTANCE_CODE(IMP_UPDATE, (IMP_UPDATE_TASK_CREATE | TASK_POLICY_TASK))) | DBG_FUNC_END,
+           task_pid(task), teffective_0(task),
+           teffective_1(task), task->priority, 0);
+
+       task_importance_update_live_donor(task);
+}
+
+
+static void
+task_policy_update_locked(task_t task, task_pend_token_t pend_token)
+{
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           (IMPORTANCE_CODE(IMP_UPDATE, TASK_POLICY_TASK) | DBG_FUNC_START),
+           task_pid(task), teffective_0(task),
+           teffective_1(task), task->priority, 0);
+
+       task_policy_update_internal_locked(task, false, pend_token);
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           (IMPORTANCE_CODE(IMP_UPDATE, TASK_POLICY_TASK)) | DBG_FUNC_END,
+           task_pid(task), teffective_0(task),
+           teffective_1(task), task->priority, 0);
+}
+
+/*
+ * One state update function TO RULE THEM ALL
+ *
+ * This function updates the task or thread effective policy fields
+ * and pushes the results to the relevant subsystems.
+ *
+ * Must call update_complete after unlocking the task,
+ * as some subsystems cannot be updated while holding the task lock.
+ *
+ * Called with task locked, not thread
+ */
+
+static void
+task_policy_update_internal_locked(task_t task, bool in_create, task_pend_token_t pend_token)
+{
+       /*
+        * Step 1:
+        *  Gather requested policy
+        */
+
+       struct task_requested_policy requested = task->requested_policy;
+
+       /*
+        * Step 2:
+        *  Calculate new effective policies from requested policy and task state
+        *  Rules:
+        *      Don't change requested, it won't take effect
+        */
+
+       struct task_effective_policy next = {};
+
+       /* Update task role */
+       next.tep_role = requested.trp_role;
+
+       /* Set task qos clamp and ceiling */
+       next.tep_qos_clamp = requested.trp_qos_clamp;
+
+       if (requested.trp_apptype == TASK_APPTYPE_APP_DEFAULT) {
+               switch (next.tep_role) {
+               case TASK_FOREGROUND_APPLICATION:
+                       /* Foreground apps get urgent scheduler priority */
+                       next.tep_qos_ui_is_urgent = 1;
+                       next.tep_qos_ceiling = THREAD_QOS_UNSPECIFIED;
+                       break;
+
+               case TASK_BACKGROUND_APPLICATION:
+                       /* This is really 'non-focal but on-screen' */
+                       next.tep_qos_ceiling = THREAD_QOS_UNSPECIFIED;
+                       break;
+
+               case TASK_DEFAULT_APPLICATION:
+                       /* This is 'may render UI but we don't know if it's focal/nonfocal' */
+                       next.tep_qos_ceiling = THREAD_QOS_UNSPECIFIED;
+                       break;
+
+               case TASK_NONUI_APPLICATION:
+                       /* i.e. 'off-screen' */
+                       next.tep_qos_ceiling = THREAD_QOS_LEGACY;
+                       break;
+
+               case TASK_CONTROL_APPLICATION:
+               case TASK_GRAPHICS_SERVER:
+                       next.tep_qos_ui_is_urgent = 1;
+                       next.tep_qos_ceiling = THREAD_QOS_UNSPECIFIED;
+                       break;
+
+               case TASK_THROTTLE_APPLICATION:
+                       /* i.e. 'TAL launch' */
+                       next.tep_qos_ceiling = THREAD_QOS_UTILITY;
+                       break;
+
+               case TASK_DARWINBG_APPLICATION:
+                       /* i.e. 'DARWIN_BG throttled background application' */
+                       next.tep_qos_ceiling = THREAD_QOS_BACKGROUND;
+                       break;
+
+               case TASK_UNSPECIFIED:
+               default:
+                       /* Apps that don't have an application role get
+                        * USER_INTERACTIVE and USER_INITIATED squashed to LEGACY */
+                       next.tep_qos_ceiling = THREAD_QOS_LEGACY;
+                       break;
+               }
+       } else {
+               /* Daemons and dext get USER_INTERACTIVE squashed to USER_INITIATED */
+               next.tep_qos_ceiling = THREAD_QOS_USER_INITIATED;
+       }
+
+       /* Calculate DARWIN_BG */
+       bool wants_darwinbg        = false;
+       bool wants_all_sockets_bg  = false; /* Do I want my existing sockets to be bg */
+       bool wants_watchersbg      = false; /* Do I want my pidbound threads to be bg */
+       bool adaptive_bg_only      = false; /* This task is BG only because it's adaptive unboosted */
+
+       /* Adaptive daemons are DARWIN_BG unless boosted, and don't get network throttled. */
+       if (requested.trp_apptype == TASK_APPTYPE_DAEMON_ADAPTIVE &&
+           requested.trp_boosted == 0) {
+               wants_darwinbg = true;
+               adaptive_bg_only = true;
+       }
+
+       /*
+        * If DARWIN_BG has been requested at either level, it's engaged.
+        * Only true DARWIN_BG changes cause watchers to transition.
+        *
+        * Backgrounding due to apptype does.
+        */
+       if (requested.trp_int_darwinbg || requested.trp_ext_darwinbg ||
+           next.tep_role == TASK_DARWINBG_APPLICATION) {
+               wants_watchersbg = wants_all_sockets_bg = wants_darwinbg = true;
+               adaptive_bg_only = false;
        }
 
-       return (KERN_SUCCESS);
+       /* Application launching in special Transparent App Lifecycle throttle mode */
+       if (requested.trp_apptype == TASK_APPTYPE_APP_DEFAULT &&
+           requested.trp_role == TASK_THROTTLE_APPLICATION) {
+               next.tep_tal_engaged = 1;
+       }
+
+       /* Background daemons are always DARWIN_BG, no exceptions, and don't get network throttled. */
+       if (requested.trp_apptype == TASK_APPTYPE_DAEMON_BACKGROUND) {
+               wants_darwinbg = true;
+               adaptive_bg_only = false;
+       }
+
+       if (next.tep_qos_clamp == THREAD_QOS_BACKGROUND ||
+           next.tep_qos_clamp == THREAD_QOS_MAINTENANCE) {
+               wants_darwinbg = true;
+               adaptive_bg_only = false;
+       }
+
+       /* Calculate side effects of DARWIN_BG */
+
+       if (wants_darwinbg) {
+               next.tep_darwinbg = 1;
+               /* darwinbg tasks always create bg sockets, but we don't always loop over all sockets */
+               next.tep_new_sockets_bg = 1;
+               next.tep_lowpri_cpu = 1;
+       }
+
+       if (wants_all_sockets_bg) {
+               next.tep_all_sockets_bg = 1;
+       }
+
+       if (wants_watchersbg) {
+               next.tep_watchers_bg = 1;
+       }
+
+       next.tep_adaptive_bg = adaptive_bg_only;
+
+       /* Calculate low CPU priority */
+
+       boolean_t wants_lowpri_cpu = false;
+
+       if (wants_darwinbg) {
+               wants_lowpri_cpu = true;
+       }
+
+       if (next.tep_tal_engaged) {
+               wants_lowpri_cpu = true;
+       }
+
+       if (requested.trp_sup_lowpri_cpu && requested.trp_boosted == 0) {
+               wants_lowpri_cpu = true;
+       }
+
+       if (wants_lowpri_cpu) {
+               next.tep_lowpri_cpu = 1;
+       }
+
+       /* Calculate IO policy */
+
+       /* Update BG IO policy (so we can see if it has changed) */
+       next.tep_bg_iotier = requested.trp_bg_iotier;
+
+       int iopol = THROTTLE_LEVEL_TIER0;
+
+       if (wants_darwinbg) {
+               iopol = MAX(iopol, requested.trp_bg_iotier);
+       }
+
+       if (requested.trp_apptype == TASK_APPTYPE_DAEMON_STANDARD) {
+               iopol = MAX(iopol, proc_standard_daemon_tier);
+       }
+
+       if (requested.trp_sup_disk && requested.trp_boosted == 0) {
+               iopol = MAX(iopol, proc_suppressed_disk_tier);
+       }
+
+       if (next.tep_tal_engaged) {
+               iopol = MAX(iopol, proc_tal_disk_tier);
+       }
+
+       if (next.tep_qos_clamp != THREAD_QOS_UNSPECIFIED) {
+               iopol = MAX(iopol, thread_qos_policy_params.qos_iotier[next.tep_qos_clamp]);
+       }
+
+       iopol = MAX(iopol, requested.trp_int_iotier);
+       iopol = MAX(iopol, requested.trp_ext_iotier);
+
+       next.tep_io_tier = iopol;
+
+       /* Calculate Passive IO policy */
+
+       if (requested.trp_ext_iopassive || requested.trp_int_iopassive) {
+               next.tep_io_passive = 1;
+       }
+
+       /* Calculate suppression-active flag */
+       boolean_t appnap_transition = false;
+
+       if (requested.trp_sup_active && requested.trp_boosted == 0) {
+               next.tep_sup_active = 1;
+       }
+
+       if (task->effective_policy.tep_sup_active != next.tep_sup_active) {
+               appnap_transition = true;
+       }
+
+       /* Calculate timer QOS */
+       int latency_qos = requested.trp_base_latency_qos;
+
+       if (requested.trp_sup_timer && requested.trp_boosted == 0) {
+               latency_qos = requested.trp_sup_timer;
+       }
+
+       if (next.tep_qos_clamp != THREAD_QOS_UNSPECIFIED) {
+               latency_qos = MAX(latency_qos, (int)thread_qos_policy_params.qos_latency_qos[next.tep_qos_clamp]);
+       }
+
+       if (requested.trp_over_latency_qos != 0) {
+               latency_qos = requested.trp_over_latency_qos;
+       }
+
+       /* Treat the windowserver special */
+       if (requested.trp_role == TASK_GRAPHICS_SERVER) {
+               latency_qos = proc_graphics_timer_qos;
+       }
+
+       next.tep_latency_qos = latency_qos;
+
+       /* Calculate throughput QOS */
+       int through_qos = requested.trp_base_through_qos;
+
+       if (requested.trp_sup_throughput && requested.trp_boosted == 0) {
+               through_qos = requested.trp_sup_throughput;
+       }
+
+       if (next.tep_qos_clamp != THREAD_QOS_UNSPECIFIED) {
+               through_qos = MAX(through_qos, (int)thread_qos_policy_params.qos_through_qos[next.tep_qos_clamp]);
+       }
+
+       if (requested.trp_over_through_qos != 0) {
+               through_qos = requested.trp_over_through_qos;
+       }
+
+       next.tep_through_qos = through_qos;
+
+       /* Calculate suppressed CPU priority */
+       if (requested.trp_sup_cpu && requested.trp_boosted == 0) {
+               next.tep_suppressed_cpu = 1;
+       }
+
+       /*
+        * Calculate background sockets
+        * Don't take into account boosting to limit transition frequency.
+        */
+       if (requested.trp_sup_bg_sockets) {
+               next.tep_all_sockets_bg = 1;
+               next.tep_new_sockets_bg = 1;
+       }
+
+       /* Apply SFI Managed class bit */
+       next.tep_sfi_managed = requested.trp_sfi_managed;
+
+       /* Calculate 'live donor' status for live importance */
+       switch (requested.trp_apptype) {
+       case TASK_APPTYPE_APP_TAL:
+       case TASK_APPTYPE_APP_DEFAULT:
+               if (requested.trp_ext_darwinbg == 1 ||
+                   (next.tep_sup_active == 1 &&
+                   (task_policy_suppression_flags & TASK_POLICY_SUPPRESSION_NONDONOR)) ||
+                   next.tep_role == TASK_DARWINBG_APPLICATION) {
+                       next.tep_live_donor = 0;
+               } else {
+                       next.tep_live_donor = 1;
+               }
+               break;
+
+       case TASK_APPTYPE_DAEMON_INTERACTIVE:
+       case TASK_APPTYPE_DAEMON_STANDARD:
+       case TASK_APPTYPE_DAEMON_ADAPTIVE:
+       case TASK_APPTYPE_DAEMON_BACKGROUND:
+       case TASK_APPTYPE_DRIVER:
+       default:
+               next.tep_live_donor = 0;
+               break;
+       }
+
+       if (requested.trp_terminated) {
+               /*
+                * Shoot down the throttles that slow down exit or response to SIGTERM
+                * We don't need to shoot down:
+                * passive        (don't want to cause others to throttle)
+                * all_sockets_bg (don't need to iterate FDs on every exit)
+                * new_sockets_bg (doesn't matter for exiting process)
+                * pidsuspend     (jetsam-ed BG process shouldn't run again)
+                * watchers_bg    (watcher threads don't need to be unthrottled)
+                * latency_qos    (affects userspace timers only)
+                */
+
+               next.tep_terminated     = 1;
+               next.tep_darwinbg       = 0;
+               next.tep_lowpri_cpu     = 0;
+               next.tep_io_tier        = THROTTLE_LEVEL_TIER0;
+               next.tep_tal_engaged    = 0;
+               next.tep_role           = TASK_UNSPECIFIED;
+               next.tep_suppressed_cpu = 0;
+       }
+
+       /*
+        * Step 3:
+        *  Swap out old policy for new policy
+        */
+
+       struct task_effective_policy prev = task->effective_policy;
+
+       /* This is the point where the new values become visible to other threads */
+       task->effective_policy = next;
+
+       /* Don't do anything further to a half-formed task */
+       if (in_create) {
+               return;
+       }
+
+       if (task == kernel_task) {
+               panic("Attempting to set task policy on kernel_task");
+       }
+
+       /*
+        * Step 4:
+        *  Pend updates that can't be done while holding the task lock
+        */
+
+       if (prev.tep_all_sockets_bg != next.tep_all_sockets_bg) {
+               pend_token->tpt_update_sockets = 1;
+       }
+
+       /* Only re-scan the timer list if the qos level is getting less strong */
+       if (prev.tep_latency_qos > next.tep_latency_qos) {
+               pend_token->tpt_update_timers = 1;
+       }
+
+#if CONFIG_TASKWATCH
+       if (prev.tep_watchers_bg != next.tep_watchers_bg) {
+               pend_token->tpt_update_watchers = 1;
+       }
+#endif /* CONFIG_TASKWATCH */
+
+       if (prev.tep_live_donor != next.tep_live_donor) {
+               pend_token->tpt_update_live_donor = 1;
+       }
+
+       /*
+        * Step 5:
+        *  Update other subsystems as necessary if something has changed
+        */
+
+       bool update_threads = false, update_sfi = false;
+
+       /*
+        * Check for the attributes that thread_policy_update_internal_locked() consults,
+        *  and trigger thread policy re-evaluation.
+        */
+       if (prev.tep_io_tier != next.tep_io_tier ||
+           prev.tep_bg_iotier != next.tep_bg_iotier ||
+           prev.tep_io_passive != next.tep_io_passive ||
+           prev.tep_darwinbg != next.tep_darwinbg ||
+           prev.tep_qos_clamp != next.tep_qos_clamp ||
+           prev.tep_qos_ceiling != next.tep_qos_ceiling ||
+           prev.tep_qos_ui_is_urgent != next.tep_qos_ui_is_urgent ||
+           prev.tep_latency_qos != next.tep_latency_qos ||
+           prev.tep_through_qos != next.tep_through_qos ||
+           prev.tep_lowpri_cpu != next.tep_lowpri_cpu ||
+           prev.tep_new_sockets_bg != next.tep_new_sockets_bg ||
+           prev.tep_terminated != next.tep_terminated ||
+           prev.tep_adaptive_bg != next.tep_adaptive_bg) {
+               update_threads = true;
+       }
+
+       /*
+        * Check for the attributes that sfi_thread_classify() consults,
+        *  and trigger SFI re-evaluation.
+        */
+       if (prev.tep_latency_qos != next.tep_latency_qos ||
+           prev.tep_role != next.tep_role ||
+           prev.tep_sfi_managed != next.tep_sfi_managed) {
+               update_sfi = true;
+       }
+
+       /* Reflect task role transitions into the coalition role counters */
+       if (prev.tep_role != next.tep_role) {
+               if (task_policy_update_coalition_focal_tasks(task, prev.tep_role, next.tep_role, pend_token)) {
+                       update_sfi = true;
+               }
+       }
+
+       bool update_priority = false;
+
+       int16_t priority     = BASEPRI_DEFAULT;
+       int16_t max_priority = MAXPRI_USER;
+
+       if (next.tep_lowpri_cpu) {
+               priority = MAXPRI_THROTTLE;
+               max_priority = MAXPRI_THROTTLE;
+       } else if (next.tep_suppressed_cpu) {
+               priority = MAXPRI_SUPPRESSED;
+               max_priority = MAXPRI_SUPPRESSED;
+       } else {
+               switch (next.tep_role) {
+               case TASK_CONTROL_APPLICATION:
+                       priority = BASEPRI_CONTROL;
+                       break;
+               case TASK_GRAPHICS_SERVER:
+                       priority = BASEPRI_GRAPHICS;
+                       max_priority = MAXPRI_RESERVED;
+                       break;
+               default:
+                       break;
+               }
+
+               /* factor in 'nice' value */
+               priority += task->importance;
+
+               if (task->effective_policy.tep_qos_clamp != THREAD_QOS_UNSPECIFIED) {
+                       int16_t qos_clamp_priority = thread_qos_policy_params.qos_pri[task->effective_policy.tep_qos_clamp];
+
+                       priority        = MIN(priority, qos_clamp_priority);
+                       max_priority    = MIN(max_priority, qos_clamp_priority);
+               }
+
+               if (priority > max_priority) {
+                       priority = max_priority;
+               } else if (priority < MINPRI) {
+                       priority = MINPRI;
+               }
+       }
+
+       assert(priority <= max_priority);
+
+       /* avoid extra work if priority isn't changing */
+       if (priority != task->priority ||
+           max_priority != task->max_priority) {
+               /* update the scheduling priority for the task */
+               task->max_priority  = max_priority;
+               task->priority      = priority;
+               update_priority     = true;
+       }
+
+       /* Loop over the threads in the task:
+        * only once
+        * only if necessary
+        * with one thread mutex hold per thread
+        */
+       if (update_threads || update_priority || update_sfi) {
+               thread_t thread;
+
+               queue_iterate(&task->threads, thread, thread_t, task_threads) {
+                       struct task_pend_token thread_pend_token = {};
+
+                       if (update_sfi) {
+                               thread_pend_token.tpt_update_thread_sfi = 1;
+                       }
+
+                       if (update_priority || update_threads) {
+                               thread_policy_update_tasklocked(thread,
+                                   task->priority, task->max_priority,
+                                   &thread_pend_token);
+                       }
+
+                       assert(!thread_pend_token.tpt_update_sockets);
+
+                       // Slightly risky, as we still hold the task lock...
+                       thread_policy_update_complete_unlocked(thread, &thread_pend_token);
+               }
+       }
+
+       /*
+        * Use the app-nap transitions to influence the
+        * transition of the process within the jetsam band
+        * [and optionally its live-donor status]
+        * On macOS only.
+        */
+       if (appnap_transition) {
+               if (task->effective_policy.tep_sup_active == 1) {
+                       memorystatus_update_priority_for_appnap(((proc_t) task->bsd_info), TRUE);
+               } else {
+                       memorystatus_update_priority_for_appnap(((proc_t) task->bsd_info), FALSE);
+               }
+       }
+}
+
+
+/*
+ * Yet another layering violation. We reach out and bang on the coalition directly.
+ */
+static boolean_t
+task_policy_update_coalition_focal_tasks(task_t            task,
+    int               prev_role,
+    int               next_role,
+    task_pend_token_t pend_token)
+{
+       boolean_t sfi_transition = FALSE;
+       uint32_t new_count = 0;
+
+       /* task moving into/out-of the foreground */
+       if (prev_role != TASK_FOREGROUND_APPLICATION && next_role == TASK_FOREGROUND_APPLICATION) {
+               if (task_coalition_adjust_focal_count(task, 1, &new_count) && (new_count == 1)) {
+                       sfi_transition = TRUE;
+                       pend_token->tpt_update_tg_ui_flag = TRUE;
+               }
+       } else if (prev_role == TASK_FOREGROUND_APPLICATION && next_role != TASK_FOREGROUND_APPLICATION) {
+               if (task_coalition_adjust_focal_count(task, -1, &new_count) && (new_count == 0)) {
+                       sfi_transition = TRUE;
+                       pend_token->tpt_update_tg_ui_flag = TRUE;
+               }
+       }
+
+       /* task moving into/out-of background */
+       if (prev_role != TASK_BACKGROUND_APPLICATION && next_role == TASK_BACKGROUND_APPLICATION) {
+               if (task_coalition_adjust_nonfocal_count(task, 1, &new_count) && (new_count == 1)) {
+                       sfi_transition = TRUE;
+               }
+       } else if (prev_role == TASK_BACKGROUND_APPLICATION && next_role != TASK_BACKGROUND_APPLICATION) {
+               if (task_coalition_adjust_nonfocal_count(task, -1, &new_count) && (new_count == 0)) {
+                       sfi_transition = TRUE;
+               }
+       }
+
+       if (sfi_transition) {
+               pend_token->tpt_update_coal_sfi = 1;
+       }
+       return sfi_transition;
+}
+
+#if CONFIG_SCHED_SFI
+
+/* coalition object is locked */
+static void
+task_sfi_reevaluate_cb(coalition_t coal, void *ctx, task_t task)
+{
+       thread_t thread;
+
+       /* unused for now */
+       (void)coal;
+
+       /* skip the task we're re-evaluating on behalf of: it's already updated */
+       if (task == (task_t)ctx) {
+               return;
+       }
+
+       task_lock(task);
+
+       queue_iterate(&task->threads, thread, thread_t, task_threads) {
+               sfi_reevaluate(thread);
+       }
+
+       task_unlock(task);
+}
+#endif /* CONFIG_SCHED_SFI */
+
+/*
+ * Called with task unlocked to do things that can't be done while holding the task lock
+ */
+void
+task_policy_update_complete_unlocked(task_t task, task_pend_token_t pend_token)
+{
+#ifdef MACH_BSD
+       if (pend_token->tpt_update_sockets) {
+               proc_apply_task_networkbg(task->bsd_info, THREAD_NULL);
+       }
+#endif /* MACH_BSD */
+
+       /* The timer throttle has been removed or reduced, we need to look for expired timers and fire them */
+       if (pend_token->tpt_update_timers) {
+               ml_timer_evaluate();
+       }
+
+#if CONFIG_TASKWATCH
+       if (pend_token->tpt_update_watchers) {
+               apply_appstate_watchers(task);
+       }
+#endif /* CONFIG_TASKWATCH */
+
+       if (pend_token->tpt_update_live_donor) {
+               task_importance_update_live_donor(task);
+       }
+
+#if CONFIG_SCHED_SFI
+       /* use the resource coalition for SFI re-evaluation */
+       if (pend_token->tpt_update_coal_sfi) {
+               coalition_for_each_task(task->coalition[COALITION_TYPE_RESOURCE],
+                   (void *)task, task_sfi_reevaluate_cb);
+       }
+#endif /* CONFIG_SCHED_SFI */
+
+#if CONFIG_THREAD_GROUPS
+       if (pend_token->tpt_update_tg_ui_flag) {
+               task_coalition_thread_group_focal_update(task);
+       }
+#endif /* CONFIG_THREAD_GROUPS */
+}
+
+/*
+ * Initiate a task policy state transition
+ *
+ * Everything that modifies requested except functions that need to hold the task lock
+ * should use this function
+ *
+ * Argument validation should be performed before reaching this point.
+ *
+ * TODO: Do we need to check task->active?
+ */
+void
+proc_set_task_policy(task_t     task,
+    int        category,
+    int        flavor,
+    int        value)
+{
+       struct task_pend_token pend_token = {};
+
+       task_lock(task);
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           (IMPORTANCE_CODE(flavor, (category | TASK_POLICY_TASK))) | DBG_FUNC_START,
+           task_pid(task), trequested_0(task),
+           trequested_1(task), value, 0);
+
+       proc_set_task_policy_locked(task, category, flavor, value, 0);
+
+       task_policy_update_locked(task, &pend_token);
+
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           (IMPORTANCE_CODE(flavor, (category | TASK_POLICY_TASK))) | DBG_FUNC_END,
+           task_pid(task), trequested_0(task),
+           trequested_1(task), tpending(&pend_token), 0);
+
+       task_unlock(task);
+
+       task_policy_update_complete_unlocked(task, &pend_token);
+}
+
+/*
+ * Variant of proc_set_task_policy() that sets two scalars in the requested policy structure.
+ * Same locking rules apply.
+ */
+void
+proc_set_task_policy2(task_t    task,
+    int       category,
+    int       flavor,
+    int       value,
+    int       value2)
+{
+       struct task_pend_token pend_token = {};
+
+       task_lock(task);
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           (IMPORTANCE_CODE(flavor, (category | TASK_POLICY_TASK))) | DBG_FUNC_START,
+           task_pid(task), trequested_0(task),
+           trequested_1(task), value, 0);
+
+       proc_set_task_policy_locked(task, category, flavor, value, value2);
+
+       task_policy_update_locked(task, &pend_token);
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           (IMPORTANCE_CODE(flavor, (category | TASK_POLICY_TASK))) | DBG_FUNC_END,
+           task_pid(task), trequested_0(task),
+           trequested_1(task), tpending(&pend_token), 0);
+
+       task_unlock(task);
+
+       task_policy_update_complete_unlocked(task, &pend_token);
+}
+
+/*
+ * Set the requested state for a specific flavor to a specific value.
+ *
+ *  TODO:
+ *  Verify that arguments to non iopol things are 1 or 0
+ */
+static void
+proc_set_task_policy_locked(task_t      task,
+    int         category,
+    int         flavor,
+    int         value,
+    int         value2)
+{
+       int tier, passive;
+
+       struct task_requested_policy requested = task->requested_policy;
+
+       switch (flavor) {
+       /* Category: EXTERNAL and INTERNAL */
+
+       case TASK_POLICY_DARWIN_BG:
+               if (category == TASK_POLICY_EXTERNAL) {
+                       requested.trp_ext_darwinbg = value;
+               } else {
+                       requested.trp_int_darwinbg = value;
+               }
+               break;
+
+       case TASK_POLICY_IOPOL:
+               proc_iopol_to_tier(value, &tier, &passive);
+               if (category == TASK_POLICY_EXTERNAL) {
+                       requested.trp_ext_iotier  = tier;
+                       requested.trp_ext_iopassive = passive;
+               } else {
+                       requested.trp_int_iotier  = tier;
+                       requested.trp_int_iopassive = passive;
+               }
+               break;
+
+       case TASK_POLICY_IO:
+               if (category == TASK_POLICY_EXTERNAL) {
+                       requested.trp_ext_iotier = value;
+               } else {
+                       requested.trp_int_iotier = value;
+               }
+               break;
+
+       case TASK_POLICY_PASSIVE_IO:
+               if (category == TASK_POLICY_EXTERNAL) {
+                       requested.trp_ext_iopassive = value;
+               } else {
+                       requested.trp_int_iopassive = value;
+               }
+               break;
+
+       /* Category: INTERNAL */
+
+       case TASK_POLICY_DARWIN_BG_IOPOL:
+               assert(category == TASK_POLICY_INTERNAL);
+               proc_iopol_to_tier(value, &tier, &passive);
+               requested.trp_bg_iotier = tier;
+               break;
+
+       /* Category: ATTRIBUTE */
+
+       case TASK_POLICY_BOOST:
+               assert(category == TASK_POLICY_ATTRIBUTE);
+               requested.trp_boosted = value;
+               break;
+
+       case TASK_POLICY_ROLE:
+               assert(category == TASK_POLICY_ATTRIBUTE);
+               requested.trp_role = value;
+               break;
+
+       case TASK_POLICY_TERMINATED:
+               assert(category == TASK_POLICY_ATTRIBUTE);
+               requested.trp_terminated = value;
+               break;
+
+       case TASK_BASE_LATENCY_QOS_POLICY:
+               assert(category == TASK_POLICY_ATTRIBUTE);
+               requested.trp_base_latency_qos = value;
+               break;
+
+       case TASK_BASE_THROUGHPUT_QOS_POLICY:
+               assert(category == TASK_POLICY_ATTRIBUTE);
+               requested.trp_base_through_qos = value;
+               break;
+
+       case TASK_POLICY_SFI_MANAGED:
+               assert(category == TASK_POLICY_ATTRIBUTE);
+               requested.trp_sfi_managed = value;
+               break;
+
+       case TASK_POLICY_BASE_LATENCY_AND_THROUGHPUT_QOS:
+               assert(category == TASK_POLICY_ATTRIBUTE);
+               requested.trp_base_latency_qos = value;
+               requested.trp_base_through_qos = value2;
+               break;
+
+       case TASK_POLICY_OVERRIDE_LATENCY_AND_THROUGHPUT_QOS:
+               assert(category == TASK_POLICY_ATTRIBUTE);
+               requested.trp_over_latency_qos = value;
+               requested.trp_over_through_qos = value2;
+               break;
+
+       default:
+               panic("unknown task policy: %d %d %d %d", category, flavor, value, value2);
+               break;
+       }
+
+       task->requested_policy = requested;
+}
+
+/*
+ * Gets what you set. Effective values may be different.
+ */
+int
+proc_get_task_policy(task_t     task,
+    int        category,
+    int        flavor)
+{
+       int value = 0;
+
+       task_lock(task);
+
+       struct task_requested_policy requested = task->requested_policy;
+
+       switch (flavor) {
+       case TASK_POLICY_DARWIN_BG:
+               if (category == TASK_POLICY_EXTERNAL) {
+                       value = requested.trp_ext_darwinbg;
+               } else {
+                       value = requested.trp_int_darwinbg;
+               }
+               break;
+       case TASK_POLICY_IOPOL:
+               if (category == TASK_POLICY_EXTERNAL) {
+                       value = proc_tier_to_iopol(requested.trp_ext_iotier,
+                           requested.trp_ext_iopassive);
+               } else {
+                       value = proc_tier_to_iopol(requested.trp_int_iotier,
+                           requested.trp_int_iopassive);
+               }
+               break;
+       case TASK_POLICY_IO:
+               if (category == TASK_POLICY_EXTERNAL) {
+                       value = requested.trp_ext_iotier;
+               } else {
+                       value = requested.trp_int_iotier;
+               }
+               break;
+       case TASK_POLICY_PASSIVE_IO:
+               if (category == TASK_POLICY_EXTERNAL) {
+                       value = requested.trp_ext_iopassive;
+               } else {
+                       value = requested.trp_int_iopassive;
+               }
+               break;
+       case TASK_POLICY_DARWIN_BG_IOPOL:
+               assert(category == TASK_POLICY_INTERNAL);
+               value = proc_tier_to_iopol(requested.trp_bg_iotier, 0);
+               break;
+       case TASK_POLICY_ROLE:
+               assert(category == TASK_POLICY_ATTRIBUTE);
+               value = requested.trp_role;
+               break;
+       case TASK_POLICY_SFI_MANAGED:
+               assert(category == TASK_POLICY_ATTRIBUTE);
+               value = requested.trp_sfi_managed;
+               break;
+       default:
+               panic("unknown policy_flavor %d", flavor);
+               break;
+       }
+
+       task_unlock(task);
+
+       return value;
+}
+
+/*
+ * Variant of proc_get_task_policy() that returns two scalar outputs.
+ */
+void
+proc_get_task_policy2(task_t task,
+    __assert_only int category,
+    int flavor,
+    int *value1,
+    int *value2)
+{
+       task_lock(task);
+
+       struct task_requested_policy requested = task->requested_policy;
+
+       switch (flavor) {
+       case TASK_POLICY_BASE_LATENCY_AND_THROUGHPUT_QOS:
+               assert(category == TASK_POLICY_ATTRIBUTE);
+               *value1 = requested.trp_base_latency_qos;
+               *value2 = requested.trp_base_through_qos;
+               break;
+
+       case TASK_POLICY_OVERRIDE_LATENCY_AND_THROUGHPUT_QOS:
+               assert(category == TASK_POLICY_ATTRIBUTE);
+               *value1 = requested.trp_over_latency_qos;
+               *value2 = requested.trp_over_through_qos;
+               break;
+
+       default:
+               panic("unknown policy_flavor %d", flavor);
+               break;
+       }
+
+       task_unlock(task);
+}
+
+/*
+ * Function for querying effective state for relevant subsystems
+ * Gets what is actually in effect, for subsystems which pull policy instead of receive updates.
+ *
+ * ONLY the relevant subsystem should query this.
+ * NEVER take a value from the 'effective' function and stuff it into a setter.
+ *
+ * NOTE: This accessor does not take the task lock.
+ * Notifications of state updates need to be externally synchronized with state queries.
+ * This routine *MUST* remain interrupt safe, as it is potentially invoked
+ * within the context of a timer interrupt.  It is also called in KDP context for stackshot.
+ */
+int
+proc_get_effective_task_policy(task_t   task,
+    int      flavor)
+{
+       int value = 0;
+
+       switch (flavor) {
+       case TASK_POLICY_DARWIN_BG:
+               /*
+                * This backs the KPI call proc_pidbackgrounded to find
+                * out if a pid is backgrounded.
+                * It is used to communicate state to the VM system, as well as
+                * prioritizing requests to the graphics system.
+                * Returns 1 for background mode, 0 for normal mode
+                */
+               value = task->effective_policy.tep_darwinbg;
+               break;
+       case TASK_POLICY_ALL_SOCKETS_BG:
+               /*
+                * do_background_socket() calls this to determine what it should do to the proc's sockets
+                * Returns 1 for background mode, 0 for normal mode
+                *
+                * This consults both thread and task so un-DBGing a thread while the task is BG
+                * doesn't get you out of the network throttle.
+                */
+               value = task->effective_policy.tep_all_sockets_bg;
+               break;
+       case TASK_POLICY_SUP_ACTIVE:
+               /*
+                * Is the task in AppNap? This is used to determine the urgency
+                * that's passed to the performance management subsystem for threads
+                * that are running at a priority <= MAXPRI_THROTTLE.
+                */
+               value = task->effective_policy.tep_sup_active;
+               break;
+       case TASK_POLICY_LATENCY_QOS:
+               /*
+                * timer arming calls into here to find out the timer coalescing level
+                * Returns a QoS tier (0-6)
+                */
+               value = task->effective_policy.tep_latency_qos;
+               break;
+       case TASK_POLICY_THROUGH_QOS:
+               /*
+                * This value is passed into the urgency callout from the scheduler
+                * to the performance management subsystem.
+                * Returns a QoS tier (0-6)
+                */
+               value = task->effective_policy.tep_through_qos;
+               break;
+       case TASK_POLICY_ROLE:
+               /*
+                * This controls various things that ask whether a process is foreground,
+                * like SFI, VM, access to GPU, etc
+                */
+               value = task->effective_policy.tep_role;
+               break;
+       case TASK_POLICY_WATCHERS_BG:
+               /*
+                * This controls whether or not a thread watching this process should be BG.
+                */
+               value = task->effective_policy.tep_watchers_bg;
+               break;
+       case TASK_POLICY_SFI_MANAGED:
+               /*
+                * This controls whether or not a process is targeted for specific control by thermald.
+                */
+               value = task->effective_policy.tep_sfi_managed;
+               break;
+       default:
+               panic("unknown policy_flavor %d", flavor);
+               break;
+       }
+
+       return value;
+}
+
+/*
+ * Convert from IOPOL_* values to throttle tiers.
+ *
+ * TODO: Can this be made more compact, like an array lookup
+ * Note that it is possible to support e.g. IOPOL_PASSIVE_STANDARD in the future
+ */
+
+void
+proc_iopol_to_tier(int iopolicy, int *tier, int *passive)
+{
+       *passive = 0;
+       *tier = 0;
+       switch (iopolicy) {
+       case IOPOL_IMPORTANT:
+               *tier = THROTTLE_LEVEL_TIER0;
+               break;
+       case IOPOL_PASSIVE:
+               *tier = THROTTLE_LEVEL_TIER0;
+               *passive = 1;
+               break;
+       case IOPOL_STANDARD:
+               *tier = THROTTLE_LEVEL_TIER1;
+               break;
+       case IOPOL_UTILITY:
+               *tier = THROTTLE_LEVEL_TIER2;
+               break;
+       case IOPOL_THROTTLE:
+               *tier = THROTTLE_LEVEL_TIER3;
+               break;
+       default:
+               panic("unknown I/O policy %d", iopolicy);
+               break;
+       }
+}
+
+int
+proc_tier_to_iopol(int tier, int passive)
+{
+       if (passive == 1) {
+               switch (tier) {
+               case THROTTLE_LEVEL_TIER0:
+                       return IOPOL_PASSIVE;
+               default:
+                       panic("unknown passive tier %d", tier);
+                       return IOPOL_DEFAULT;
+               }
+       } else {
+               switch (tier) {
+               case THROTTLE_LEVEL_NONE:
+               case THROTTLE_LEVEL_TIER0:
+                       return IOPOL_DEFAULT;
+               case THROTTLE_LEVEL_TIER1:
+                       return IOPOL_STANDARD;
+               case THROTTLE_LEVEL_TIER2:
+                       return IOPOL_UTILITY;
+               case THROTTLE_LEVEL_TIER3:
+                       return IOPOL_THROTTLE;
+               default:
+                       panic("unknown tier %d", tier);
+                       return IOPOL_DEFAULT;
+               }
+       }
+}
+
+int
+proc_darwin_role_to_task_role(int darwin_role, task_role_t* task_role)
+{
+       integer_t role = TASK_UNSPECIFIED;
+
+       switch (darwin_role) {
+       case PRIO_DARWIN_ROLE_DEFAULT:
+               role = TASK_UNSPECIFIED;
+               break;
+       case PRIO_DARWIN_ROLE_UI_FOCAL:
+               role = TASK_FOREGROUND_APPLICATION;
+               break;
+       case PRIO_DARWIN_ROLE_UI:
+               role = TASK_DEFAULT_APPLICATION;
+               break;
+       case PRIO_DARWIN_ROLE_NON_UI:
+               role = TASK_NONUI_APPLICATION;
+               break;
+       case PRIO_DARWIN_ROLE_UI_NON_FOCAL:
+               role = TASK_BACKGROUND_APPLICATION;
+               break;
+       case PRIO_DARWIN_ROLE_TAL_LAUNCH:
+               role = TASK_THROTTLE_APPLICATION;
+               break;
+       case PRIO_DARWIN_ROLE_DARWIN_BG:
+               role = TASK_DARWINBG_APPLICATION;
+               break;
+       default:
+               return EINVAL;
+       }
+
+       *task_role = role;
+
+       return 0;
+}
+
+int
+proc_task_role_to_darwin_role(task_role_t task_role)
+{
+       switch (task_role) {
+       case TASK_FOREGROUND_APPLICATION:
+               return PRIO_DARWIN_ROLE_UI_FOCAL;
+       case TASK_BACKGROUND_APPLICATION:
+               return PRIO_DARWIN_ROLE_UI_NON_FOCAL;
+       case TASK_NONUI_APPLICATION:
+               return PRIO_DARWIN_ROLE_NON_UI;
+       case TASK_DEFAULT_APPLICATION:
+               return PRIO_DARWIN_ROLE_UI;
+       case TASK_THROTTLE_APPLICATION:
+               return PRIO_DARWIN_ROLE_TAL_LAUNCH;
+       case TASK_DARWINBG_APPLICATION:
+               return PRIO_DARWIN_ROLE_DARWIN_BG;
+       case TASK_UNSPECIFIED:
+       default:
+               return PRIO_DARWIN_ROLE_DEFAULT;
+       }
+}
+
+
+/* TODO: remove this variable when interactive daemon audit period is over */
+static TUNABLE(bool, ipc_importance_interactive_receiver,
+    "imp_interactive_receiver", false);
+
+/*
+ * Called at process exec to initialize the apptype, qos clamp, and qos seed of a process
+ *
+ * TODO: Make this function more table-driven instead of ad-hoc
+ */
+void
+proc_set_task_spawnpolicy(task_t task, thread_t thread, int apptype, int qos_clamp, task_role_t role,
+    ipc_port_t * portwatch_ports, uint32_t portwatch_count)
+{
+       struct task_pend_token pend_token = {};
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           (IMPORTANCE_CODE(IMP_TASK_APPTYPE, apptype)) | DBG_FUNC_START,
+           task_pid(task), trequested_0(task), trequested_1(task),
+           apptype, 0);
+
+       switch (apptype) {
+       case TASK_APPTYPE_APP_DEFAULT:
+               /* Apps become donors via the 'live-donor' flag instead of the static donor flag */
+               task_importance_mark_donor(task, FALSE);
+               task_importance_mark_live_donor(task, TRUE);
+               task_importance_mark_receiver(task, FALSE);
+#if !defined(XNU_TARGET_OS_OSX)
+               task_importance_mark_denap_receiver(task, FALSE);
+#else
+               /* Apps are de-nap recievers on macOS for suppression behaviors */
+               task_importance_mark_denap_receiver(task, TRUE);
+#endif /* !defined(XNU_TARGET_OS_OSX) */
+               break;
+
+       case TASK_APPTYPE_DAEMON_INTERACTIVE:
+               task_importance_mark_donor(task, TRUE);
+               task_importance_mark_live_donor(task, FALSE);
+
+               /*
+                * A boot arg controls whether interactive daemons are importance receivers.
+                * Normally, they are not.  But for testing their behavior as an adaptive
+                * daemon, the boot-arg can be set.
+                *
+                * TODO: remove this when the interactive daemon audit period is over.
+                */
+               task_importance_mark_receiver(task, /* FALSE */ ipc_importance_interactive_receiver);
+               task_importance_mark_denap_receiver(task, FALSE);
+               break;
+
+       case TASK_APPTYPE_DAEMON_STANDARD:
+               task_importance_mark_donor(task, TRUE);
+               task_importance_mark_live_donor(task, FALSE);
+               task_importance_mark_receiver(task, FALSE);
+               task_importance_mark_denap_receiver(task, FALSE);
+               break;
+
+       case TASK_APPTYPE_DAEMON_ADAPTIVE:
+               task_importance_mark_donor(task, FALSE);
+               task_importance_mark_live_donor(task, FALSE);
+               task_importance_mark_receiver(task, TRUE);
+               task_importance_mark_denap_receiver(task, FALSE);
+               break;
+
+       case TASK_APPTYPE_DAEMON_BACKGROUND:
+               task_importance_mark_donor(task, FALSE);
+               task_importance_mark_live_donor(task, FALSE);
+               task_importance_mark_receiver(task, FALSE);
+               task_importance_mark_denap_receiver(task, FALSE);
+               break;
+
+       case TASK_APPTYPE_DRIVER:
+               task_importance_mark_donor(task, FALSE);
+               task_importance_mark_live_donor(task, FALSE);
+               task_importance_mark_receiver(task, FALSE);
+               task_importance_mark_denap_receiver(task, FALSE);
+               break;
+
+       case TASK_APPTYPE_NONE:
+               break;
+       }
+
+       if (portwatch_ports != NULL && apptype == TASK_APPTYPE_DAEMON_ADAPTIVE) {
+               int portwatch_boosts = 0;
+
+               for (uint32_t i = 0; i < portwatch_count; i++) {
+                       ipc_port_t port = NULL;
+
+                       if (IP_VALID(port = portwatch_ports[i])) {
+                               int boost = 0;
+                               task_add_importance_watchport(task, port, &boost);
+                               portwatch_boosts += boost;
+                       }
+               }
+
+               if (portwatch_boosts > 0) {
+                       task_importance_hold_internal_assertion(task, portwatch_boosts);
+               }
+       }
+
+       /* Redirect the turnstile push of watchports to task */
+       if (portwatch_count && portwatch_ports != NULL) {
+               task_add_turnstile_watchports(task, thread, portwatch_ports, portwatch_count);
+       }
+
+       task_lock(task);
+
+       if (apptype != TASK_APPTYPE_NONE) {
+               task->requested_policy.trp_apptype = apptype;
+       }
+
+#if !defined(XNU_TARGET_OS_OSX)
+       /* Remove this after launchd starts setting it properly */
+       if (apptype == TASK_APPTYPE_APP_DEFAULT && role == TASK_UNSPECIFIED) {
+               task->requested_policy.trp_role = TASK_FOREGROUND_APPLICATION;
+       } else
+#endif
+       if (role != TASK_UNSPECIFIED) {
+               task->requested_policy.trp_role = (uint32_t)role;
+       }
+
+       if (qos_clamp != THREAD_QOS_UNSPECIFIED) {
+               task->requested_policy.trp_qos_clamp = qos_clamp;
+       }
+
+       task_policy_update_locked(task, &pend_token);
+
+       task_unlock(task);
+
+       /* Ensure the donor bit is updated to be in sync with the new live donor status */
+       pend_token.tpt_update_live_donor = 1;
+
+       task_policy_update_complete_unlocked(task, &pend_token);
+
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+           (IMPORTANCE_CODE(IMP_TASK_APPTYPE, apptype)) | DBG_FUNC_END,
+           task_pid(task), trequested_0(task), trequested_1(task),
+           task_is_importance_receiver(task), 0);
+}
+
+/*
+ * Inherit task role across exec
+ */
+void
+proc_inherit_task_role(task_t new_task,
+    task_t old_task)
+{
+       int role;
+
+       /* inherit the role from old task to new task */
+       role = proc_get_task_policy(old_task, TASK_POLICY_ATTRIBUTE, TASK_POLICY_ROLE);
+       proc_set_task_policy(new_task, TASK_POLICY_ATTRIBUTE, TASK_POLICY_ROLE, role);
+}
+
+extern void * XNU_PTRAUTH_SIGNED_PTR("initproc") initproc;
+
+/*
+ * Compute the default main thread qos for a task
+ */
+thread_qos_t
+task_compute_main_thread_qos(task_t task)
+{
+       thread_qos_t primordial_qos = THREAD_QOS_UNSPECIFIED;
+
+       thread_qos_t qos_clamp = task->requested_policy.trp_qos_clamp;
+
+       switch (task->requested_policy.trp_apptype) {
+       case TASK_APPTYPE_APP_TAL:
+       case TASK_APPTYPE_APP_DEFAULT:
+               primordial_qos = THREAD_QOS_USER_INTERACTIVE;
+               break;
+
+       case TASK_APPTYPE_DAEMON_INTERACTIVE:
+       case TASK_APPTYPE_DAEMON_STANDARD:
+       case TASK_APPTYPE_DAEMON_ADAPTIVE:
+       case TASK_APPTYPE_DRIVER:
+               primordial_qos = THREAD_QOS_LEGACY;
+               break;
+
+       case TASK_APPTYPE_DAEMON_BACKGROUND:
+               primordial_qos = THREAD_QOS_BACKGROUND;
+               break;
+       }
+
+       if (task->bsd_info == initproc) {
+               /* PID 1 gets a special case */
+               primordial_qos = MAX(primordial_qos, THREAD_QOS_USER_INITIATED);
+       }
+
+       if (qos_clamp != THREAD_QOS_UNSPECIFIED) {
+               if (primordial_qos != THREAD_QOS_UNSPECIFIED) {
+                       primordial_qos = MIN(qos_clamp, primordial_qos);
+               } else {
+                       primordial_qos = qos_clamp;
+               }
+       }
+
+       return primordial_qos;
+}
+
+
+/* for process_policy to check before attempting to set */
+boolean_t
+proc_task_is_tal(task_t task)
+{
+       return (task->requested_policy.trp_apptype == TASK_APPTYPE_APP_TAL) ? TRUE : FALSE;
+}
+
+int
+task_get_apptype(task_t task)
+{
+       return task->requested_policy.trp_apptype;
+}
+
+boolean_t
+task_is_daemon(task_t task)
+{
+       switch (task->requested_policy.trp_apptype) {
+       case TASK_APPTYPE_DAEMON_INTERACTIVE:
+       case TASK_APPTYPE_DAEMON_STANDARD:
+       case TASK_APPTYPE_DAEMON_ADAPTIVE:
+       case TASK_APPTYPE_DAEMON_BACKGROUND:
+               return TRUE;
+       default:
+               return FALSE;
+       }
+}
+
+bool
+task_is_driver(task_t task)
+{
+       if (!task) {
+               return FALSE;
+       }
+       return task->requested_policy.trp_apptype == TASK_APPTYPE_DRIVER;
+}
+
+boolean_t
+task_is_app(task_t task)
+{
+       switch (task->requested_policy.trp_apptype) {
+       case TASK_APPTYPE_APP_DEFAULT:
+       case TASK_APPTYPE_APP_TAL:
+               return TRUE;
+       default:
+               return FALSE;
+       }
+}
+
+/* for telemetry */
+integer_t
+task_grab_latency_qos(task_t task)
+{
+       return qos_latency_policy_package(proc_get_effective_task_policy(task, TASK_POLICY_LATENCY_QOS));
+}
+
+/* update the darwin background action state in the flags field for libproc */
+int
+proc_get_darwinbgstate(task_t task, uint32_t * flagsp)
+{
+       if (task->requested_policy.trp_ext_darwinbg) {
+               *flagsp |= PROC_FLAG_EXT_DARWINBG;
+       }
+
+       if (task->requested_policy.trp_int_darwinbg) {
+               *flagsp |= PROC_FLAG_DARWINBG;
+       }
+
+#if !defined(XNU_TARGET_OS_OSX)
+       if (task->requested_policy.trp_apptype == TASK_APPTYPE_DAEMON_BACKGROUND) {
+               *flagsp |= PROC_FLAG_IOS_APPLEDAEMON;
+       }
+
+       if (task->requested_policy.trp_apptype == TASK_APPTYPE_DAEMON_ADAPTIVE) {
+               *flagsp |= PROC_FLAG_IOS_IMPPROMOTION;
+       }
+#endif /* !defined(XNU_TARGET_OS_OSX) */
+
+       if (task->requested_policy.trp_apptype == TASK_APPTYPE_APP_DEFAULT ||
+           task->requested_policy.trp_apptype == TASK_APPTYPE_APP_TAL) {
+               *flagsp |= PROC_FLAG_APPLICATION;
+       }
+
+       if (task->requested_policy.trp_apptype == TASK_APPTYPE_DAEMON_ADAPTIVE) {
+               *flagsp |= PROC_FLAG_ADAPTIVE;
+       }
+
+       if (task->requested_policy.trp_apptype == TASK_APPTYPE_DAEMON_ADAPTIVE &&
+           task->requested_policy.trp_boosted == 1) {
+               *flagsp |= PROC_FLAG_ADAPTIVE_IMPORTANT;
+       }
+
+       if (task_is_importance_donor(task)) {
+               *flagsp |= PROC_FLAG_IMPORTANCE_DONOR;
+       }
+
+       if (task->effective_policy.tep_sup_active) {
+               *flagsp |= PROC_FLAG_SUPPRESSED;
+       }
+
+       return 0;
+}
+
+/*
+ * Tracepoint data... Reading the tracepoint data can be somewhat complicated.
+ * The current scheme packs as much data into a single tracepoint as it can.
+ *
+ * Each task/thread requested/effective structure is 64 bits in size. Any
+ * given tracepoint will emit either requested or effective data, but not both.
+ *
+ * A tracepoint may emit any of task, thread, or task & thread data.
+ *
+ * The type of data emitted varies with pointer size. Where possible, both
+ * task and thread data are emitted. In LP32 systems, the first and second
+ * halves of either the task or thread data is emitted.
+ *
+ * The code uses uintptr_t array indexes instead of high/low to avoid
+ * confusion WRT big vs little endian.
+ *
+ * The truth table for the tracepoint data functions is below, and has the
+ * following invariants:
+ *
+ * 1) task and thread are uintptr_t*
+ * 2) task may never be NULL
+ *
+ *
+ *                                     LP32            LP64
+ * trequested_0(task, NULL)            task[0]         task[0]
+ * trequested_1(task, NULL)            task[1]         NULL
+ * trequested_0(task, thread)          thread[0]       task[0]
+ * trequested_1(task, thread)          thread[1]       thread[0]
+ *
+ * Basically, you get a full task or thread on LP32, and both on LP64.
+ *
+ * The uintptr_t munging here is squicky enough to deserve a comment.
+ *
+ * The variables we are accessing are laid out in memory like this:
+ *
+ * [            LP64 uintptr_t  0          ]
+ * [ LP32 uintptr_t 0 ] [ LP32 uintptr_t 1 ]
+ *
+ *      1   2   3   4     5   6   7   8
+ *
+ */
+
+static uintptr_t
+trequested_0(task_t task)
+{
+       static_assert(sizeof(struct task_requested_policy) == sizeof(uint64_t), "size invariant violated");
+
+       uintptr_t* raw = (uintptr_t*)&task->requested_policy;
+
+       return raw[0];
+}
+
+static uintptr_t
+trequested_1(task_t task)
+{
+#if defined __LP64__
+       (void)task;
+       return 0;
+#else
+       uintptr_t* raw = (uintptr_t*)(&task->requested_policy);
+       return raw[1];
+#endif
+}
+
+static uintptr_t
+teffective_0(task_t task)
+{
+       uintptr_t* raw = (uintptr_t*)&task->effective_policy;
+
+       return raw[0];
+}
+
+static uintptr_t
+teffective_1(task_t task)
+{
+#if defined __LP64__
+       (void)task;
+       return 0;
+#else
+       uintptr_t* raw = (uintptr_t*)(&task->effective_policy);
+       return raw[1];
+#endif
+}
+
+/* dump pending for tracepoint */
+uint32_t
+tpending(task_pend_token_t pend_token)
+{
+       return *(uint32_t*)(void*)(pend_token);
+}
+
+uint64_t
+task_requested_bitfield(task_t task)
+{
+       uint64_t bits = 0;
+       struct task_requested_policy requested = task->requested_policy;
+
+       bits |= (requested.trp_int_darwinbg     ? POLICY_REQ_INT_DARWIN_BG  : 0);
+       bits |= (requested.trp_ext_darwinbg     ? POLICY_REQ_EXT_DARWIN_BG  : 0);
+       bits |= (requested.trp_int_iotier       ? (((uint64_t)requested.trp_int_iotier) << POLICY_REQ_INT_IO_TIER_SHIFT) : 0);
+       bits |= (requested.trp_ext_iotier       ? (((uint64_t)requested.trp_ext_iotier) << POLICY_REQ_EXT_IO_TIER_SHIFT) : 0);
+       bits |= (requested.trp_int_iopassive    ? POLICY_REQ_INT_PASSIVE_IO : 0);
+       bits |= (requested.trp_ext_iopassive    ? POLICY_REQ_EXT_PASSIVE_IO : 0);
+       bits |= (requested.trp_bg_iotier        ? (((uint64_t)requested.trp_bg_iotier) << POLICY_REQ_BG_IOTIER_SHIFT)   : 0);
+       bits |= (requested.trp_terminated       ? POLICY_REQ_TERMINATED     : 0);
+
+       bits |= (requested.trp_boosted          ? POLICY_REQ_BOOSTED        : 0);
+       bits |= (requested.trp_tal_enabled      ? POLICY_REQ_TAL_ENABLED    : 0);
+       bits |= (requested.trp_apptype          ? (((uint64_t)requested.trp_apptype) << POLICY_REQ_APPTYPE_SHIFT)  : 0);
+       bits |= (requested.trp_role             ? (((uint64_t)requested.trp_role) << POLICY_REQ_ROLE_SHIFT)     : 0);
+
+       bits |= (requested.trp_sup_active       ? POLICY_REQ_SUP_ACTIVE         : 0);
+       bits |= (requested.trp_sup_lowpri_cpu   ? POLICY_REQ_SUP_LOWPRI_CPU     : 0);
+       bits |= (requested.trp_sup_cpu          ? POLICY_REQ_SUP_CPU            : 0);
+       bits |= (requested.trp_sup_timer        ? (((uint64_t)requested.trp_sup_timer) << POLICY_REQ_SUP_TIMER_THROTTLE_SHIFT) : 0);
+       bits |= (requested.trp_sup_throughput   ? (((uint64_t)requested.trp_sup_throughput) << POLICY_REQ_SUP_THROUGHPUT_SHIFT)     : 0);
+       bits |= (requested.trp_sup_disk         ? POLICY_REQ_SUP_DISK_THROTTLE  : 0);
+       bits |= (requested.trp_sup_bg_sockets   ? POLICY_REQ_SUP_BG_SOCKETS     : 0);
+
+       bits |= (requested.trp_base_latency_qos ? (((uint64_t)requested.trp_base_latency_qos) << POLICY_REQ_BASE_LATENCY_QOS_SHIFT) : 0);
+       bits |= (requested.trp_over_latency_qos ? (((uint64_t)requested.trp_over_latency_qos) << POLICY_REQ_OVER_LATENCY_QOS_SHIFT) : 0);
+       bits |= (requested.trp_base_through_qos ? (((uint64_t)requested.trp_base_through_qos) << POLICY_REQ_BASE_THROUGH_QOS_SHIFT) : 0);
+       bits |= (requested.trp_over_through_qos ? (((uint64_t)requested.trp_over_through_qos) << POLICY_REQ_OVER_THROUGH_QOS_SHIFT) : 0);
+       bits |= (requested.trp_sfi_managed      ? POLICY_REQ_SFI_MANAGED        : 0);
+       bits |= (requested.trp_qos_clamp        ? (((uint64_t)requested.trp_qos_clamp) << POLICY_REQ_QOS_CLAMP_SHIFT)        : 0);
+
+       return bits;
+}
+
+uint64_t
+task_effective_bitfield(task_t task)
+{
+       uint64_t bits = 0;
+       struct task_effective_policy effective = task->effective_policy;
+
+       bits |= (effective.tep_io_tier          ? (((uint64_t)effective.tep_io_tier) << POLICY_EFF_IO_TIER_SHIFT) : 0);
+       bits |= (effective.tep_io_passive       ? POLICY_EFF_IO_PASSIVE     : 0);
+       bits |= (effective.tep_darwinbg         ? POLICY_EFF_DARWIN_BG      : 0);
+       bits |= (effective.tep_lowpri_cpu       ? POLICY_EFF_LOWPRI_CPU     : 0);
+       bits |= (effective.tep_terminated       ? POLICY_EFF_TERMINATED     : 0);
+       bits |= (effective.tep_all_sockets_bg   ? POLICY_EFF_ALL_SOCKETS_BG : 0);
+       bits |= (effective.tep_new_sockets_bg   ? POLICY_EFF_NEW_SOCKETS_BG : 0);
+       bits |= (effective.tep_bg_iotier        ? (((uint64_t)effective.tep_bg_iotier) << POLICY_EFF_BG_IOTIER_SHIFT) : 0);
+       bits |= (effective.tep_qos_ui_is_urgent ? POLICY_EFF_QOS_UI_IS_URGENT : 0);
+
+       bits |= (effective.tep_tal_engaged      ? POLICY_EFF_TAL_ENGAGED    : 0);
+       bits |= (effective.tep_watchers_bg      ? POLICY_EFF_WATCHERS_BG    : 0);
+       bits |= (effective.tep_sup_active       ? POLICY_EFF_SUP_ACTIVE     : 0);
+       bits |= (effective.tep_suppressed_cpu   ? POLICY_EFF_SUP_CPU        : 0);
+       bits |= (effective.tep_role             ? (((uint64_t)effective.tep_role) << POLICY_EFF_ROLE_SHIFT)        : 0);
+       bits |= (effective.tep_latency_qos      ? (((uint64_t)effective.tep_latency_qos) << POLICY_EFF_LATENCY_QOS_SHIFT) : 0);
+       bits |= (effective.tep_through_qos      ? (((uint64_t)effective.tep_through_qos) << POLICY_EFF_THROUGH_QOS_SHIFT) : 0);
+       bits |= (effective.tep_sfi_managed      ? POLICY_EFF_SFI_MANAGED    : 0);
+       bits |= (effective.tep_qos_ceiling      ? (((uint64_t)effective.tep_qos_ceiling) << POLICY_EFF_QOS_CEILING_SHIFT) : 0);
+
+       return bits;
+}
+
+
+/*
+ * Resource usage and CPU related routines
+ */
+
+int
+proc_get_task_ruse_cpu(task_t task, uint32_t *policyp, uint8_t *percentagep, uint64_t *intervalp, uint64_t *deadlinep)
+{
+       int error = 0;
+       int scope;
+
+       task_lock(task);
+
+
+       error = task_get_cpuusage(task, percentagep, intervalp, deadlinep, &scope);
+       task_unlock(task);
+
+       /*
+        * Reverse-map from CPU resource limit scopes back to policies (see comment below).
+        */
+       if (scope == TASK_RUSECPU_FLAGS_PERTHR_LIMIT) {
+               *policyp = TASK_POLICY_RESOURCE_ATTRIBUTE_NOTIFY_EXC;
+       } else if (scope == TASK_RUSECPU_FLAGS_PROC_LIMIT) {
+               *policyp = TASK_POLICY_RESOURCE_ATTRIBUTE_THROTTLE;
+       } else if (scope == TASK_RUSECPU_FLAGS_DEADLINE) {
+               *policyp = TASK_POLICY_RESOURCE_ATTRIBUTE_NONE;
+       }
+
+       return error;
+}
+
+/*
+ * Configure the default CPU usage monitor parameters.
+ *
+ * For tasks which have this mechanism activated: if any thread in the
+ * process consumes more CPU than this, an EXC_RESOURCE exception will be generated.
+ */
+void
+proc_init_cpumon_params(void)
+{
+       /*
+        * The max CPU percentage can be configured via the boot-args and
+        * a key in the device tree. The boot-args are honored first, then the
+        * device tree.
+        */
+       if (!PE_parse_boot_argn("max_cpumon_percentage", &proc_max_cpumon_percentage,
+           sizeof(proc_max_cpumon_percentage))) {
+               uint64_t max_percentage = 0ULL;
+
+               if (!PE_get_default("kern.max_cpumon_percentage", &max_percentage,
+                   sizeof(max_percentage))) {
+                       max_percentage = DEFAULT_CPUMON_PERCENTAGE;
+               }
+
+               assert(max_percentage <= UINT8_MAX);
+               proc_max_cpumon_percentage = (uint8_t) max_percentage;
+       }
+
+       if (proc_max_cpumon_percentage > 100) {
+               proc_max_cpumon_percentage = 100;
+       }
+
+       /*
+        * The interval should be specified in seconds.
+        *
+        * Like the max CPU percentage, the max CPU interval can be configured
+        * via boot-args and the device tree.
+        */
+       if (!PE_parse_boot_argn("max_cpumon_interval", &proc_max_cpumon_interval,
+           sizeof(proc_max_cpumon_interval))) {
+               if (!PE_get_default("kern.max_cpumon_interval", &proc_max_cpumon_interval,
+                   sizeof(proc_max_cpumon_interval))) {
+                       proc_max_cpumon_interval = DEFAULT_CPUMON_INTERVAL;
+               }
+       }
+
+       proc_max_cpumon_interval *= NSEC_PER_SEC;
+
+       /* TEMPORARY boot arg to control App suppression */
+       PE_parse_boot_argn("task_policy_suppression_flags",
+           &task_policy_suppression_flags,
+           sizeof(task_policy_suppression_flags));
+
+       /* adjust suppression disk policy if called for in boot arg */
+       if (task_policy_suppression_flags & TASK_POLICY_SUPPRESSION_IOTIER2) {
+               proc_suppressed_disk_tier = THROTTLE_LEVEL_TIER2;
+       }
+}
+
+/*
+ * Currently supported configurations for CPU limits.
+ *
+ * Policy                              | Deadline-based CPU limit | Percentage-based CPU limit
+ * -------------------------------------+--------------------------+------------------------------
+ * PROC_POLICY_RSRCACT_THROTTLE                | ENOTSUP                  | Task-wide scope only
+ * PROC_POLICY_RSRCACT_SUSPEND         | Task-wide scope only     | ENOTSUP
+ * PROC_POLICY_RSRCACT_TERMINATE       | Task-wide scope only     | ENOTSUP
+ * PROC_POLICY_RSRCACT_NOTIFY_KQ       | Task-wide scope only     | ENOTSUP
+ * PROC_POLICY_RSRCACT_NOTIFY_EXC      | ENOTSUP                  | Per-thread scope only
+ *
+ * A deadline-based CPU limit is actually a simple wallclock timer - the requested action is performed
+ * after the specified amount of wallclock time has elapsed.
+ *
+ * A percentage-based CPU limit performs the requested action after the specified amount of actual CPU time
+ * has been consumed -- regardless of how much wallclock time has elapsed -- by either the task as an
+ * aggregate entity (so-called "Task-wide" or "Proc-wide" scope, whereby the CPU time consumed by all threads
+ * in the task are added together), or by any one thread in the task (so-called "per-thread" scope).
+ *
+ * We support either deadline != 0 OR percentage != 0, but not both. The original intention in having them
+ * share an API was to use actual CPU time as the basis of the deadline-based limit (as in: perform an action
+ * after I have used some amount of CPU time; this is different than the recurring percentage/interval model)
+ * but the potential consumer of the API at the time was insisting on wallclock time instead.
+ *
+ * Currently, requesting notification via an exception is the only way to get per-thread scope for a
+ * CPU limit. All other types of notifications force task-wide scope for the limit.
+ */
+int
+proc_set_task_ruse_cpu(task_t task, uint16_t policy, uint8_t percentage, uint64_t interval, uint64_t deadline,
+    int cpumon_entitled)
+{
+       int error = 0;
+       int scope;
+
+       /*
+        * Enforce the matrix of supported configurations for policy, percentage, and deadline.
+        */
+       switch (policy) {
+       // If no policy is explicitly given, the default is to throttle.
+       case TASK_POLICY_RESOURCE_ATTRIBUTE_NONE:
+       case TASK_POLICY_RESOURCE_ATTRIBUTE_THROTTLE:
+               if (deadline != 0) {
+                       return ENOTSUP;
+               }
+               scope = TASK_RUSECPU_FLAGS_PROC_LIMIT;
+               break;
+       case TASK_POLICY_RESOURCE_ATTRIBUTE_SUSPEND:
+       case TASK_POLICY_RESOURCE_ATTRIBUTE_TERMINATE:
+       case TASK_POLICY_RESOURCE_ATTRIBUTE_NOTIFY_KQ:
+               if (percentage != 0) {
+                       return ENOTSUP;
+               }
+               scope = TASK_RUSECPU_FLAGS_DEADLINE;
+               break;
+       case TASK_POLICY_RESOURCE_ATTRIBUTE_NOTIFY_EXC:
+               if (deadline != 0) {
+                       return ENOTSUP;
+               }
+               scope = TASK_RUSECPU_FLAGS_PERTHR_LIMIT;
+#ifdef CONFIG_NOMONITORS
+               return error;
+#endif /* CONFIG_NOMONITORS */
+               break;
+       default:
+               return EINVAL;
+       }
+
+       task_lock(task);
+       if (task != current_task()) {
+               task->policy_ru_cpu_ext = policy;
+       } else {
+               task->policy_ru_cpu = policy;
+       }
+       error = task_set_cpuusage(task, percentage, interval, deadline, scope, cpumon_entitled);
+       task_unlock(task);
+       return error;
+}
+
+/* TODO: get rid of these */
+#define TASK_POLICY_CPU_RESOURCE_USAGE          0
+#define TASK_POLICY_WIREDMEM_RESOURCE_USAGE     1
+#define TASK_POLICY_VIRTUALMEM_RESOURCE_USAGE   2
+#define TASK_POLICY_DISK_RESOURCE_USAGE         3
+#define TASK_POLICY_NETWORK_RESOURCE_USAGE      4
+#define TASK_POLICY_POWER_RESOURCE_USAGE        5
+
+#define TASK_POLICY_RESOURCE_USAGE_COUNT        6
+
+int
+proc_clear_task_ruse_cpu(task_t task, int cpumon_entitled)
+{
+       int error = 0;
+       int action;
+       void * bsdinfo = NULL;
+
+       task_lock(task);
+       if (task != current_task()) {
+               task->policy_ru_cpu_ext = TASK_POLICY_RESOURCE_ATTRIBUTE_DEFAULT;
+       } else {
+               task->policy_ru_cpu = TASK_POLICY_RESOURCE_ATTRIBUTE_DEFAULT;
+       }
+
+       error = task_clear_cpuusage_locked(task, cpumon_entitled);
+       if (error != 0) {
+               goto out;
+       }
+
+       action = task->applied_ru_cpu;
+       if (task->applied_ru_cpu_ext != TASK_POLICY_RESOURCE_ATTRIBUTE_NONE) {
+               /* reset action */
+               task->applied_ru_cpu_ext = TASK_POLICY_RESOURCE_ATTRIBUTE_NONE;
+       }
+       if (action != TASK_POLICY_RESOURCE_ATTRIBUTE_NONE) {
+               bsdinfo = task->bsd_info;
+               task_unlock(task);
+               proc_restore_resource_actions(bsdinfo, TASK_POLICY_CPU_RESOURCE_USAGE, action);
+               goto out1;
+       }
+
+out:
+       task_unlock(task);
+out1:
+       return error;
+}
+
+/* used to apply resource limit related actions */
+static int
+task_apply_resource_actions(task_t task, int type)
+{
+       int action = TASK_POLICY_RESOURCE_ATTRIBUTE_NONE;
+       void * bsdinfo = NULL;
+
+       switch (type) {
+       case TASK_POLICY_CPU_RESOURCE_USAGE:
+               break;
+       case TASK_POLICY_WIREDMEM_RESOURCE_USAGE:
+       case TASK_POLICY_VIRTUALMEM_RESOURCE_USAGE:
+       case TASK_POLICY_DISK_RESOURCE_USAGE:
+       case TASK_POLICY_NETWORK_RESOURCE_USAGE:
+       case TASK_POLICY_POWER_RESOURCE_USAGE:
+               return 0;
+
+       default:
+               return 1;
+       }
+       ;
+
+       /* only cpu actions for now */
+       task_lock(task);
+
+       if (task->applied_ru_cpu_ext == TASK_POLICY_RESOURCE_ATTRIBUTE_NONE) {
+               /* apply action */
+               task->applied_ru_cpu_ext = task->policy_ru_cpu_ext;
+               action = task->applied_ru_cpu_ext;
+       } else {
+               action = task->applied_ru_cpu_ext;
+       }
+
+       if (action != TASK_POLICY_RESOURCE_ATTRIBUTE_NONE) {
+               bsdinfo = task->bsd_info;
+               task_unlock(task);
+               proc_apply_resource_actions(bsdinfo, TASK_POLICY_CPU_RESOURCE_USAGE, action);
+       } else {
+               task_unlock(task);
+       }
+
+       return 0;
+}
+
+/*
+ * XXX This API is somewhat broken; we support multiple simultaneous CPU limits, but the get/set API
+ * only allows for one at a time. This means that if there is a per-thread limit active, the other
+ * "scopes" will not be accessible via this API. We could change it to pass in the scope of interest
+ * to the caller, and prefer that, but there's no need for that at the moment.
+ */
+static int
+task_get_cpuusage(task_t task, uint8_t *percentagep, uint64_t *intervalp, uint64_t *deadlinep, int *scope)
+{
+       *percentagep = 0;
+       *intervalp = 0;
+       *deadlinep = 0;
+
+       if ((task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_PERTHR_LIMIT) != 0) {
+               *scope = TASK_RUSECPU_FLAGS_PERTHR_LIMIT;
+               *percentagep = task->rusage_cpu_perthr_percentage;
+               *intervalp = task->rusage_cpu_perthr_interval;
+       } else if ((task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_PROC_LIMIT) != 0) {
+               *scope = TASK_RUSECPU_FLAGS_PROC_LIMIT;
+               *percentagep = task->rusage_cpu_percentage;
+               *intervalp = task->rusage_cpu_interval;
+       } else if ((task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_DEADLINE) != 0) {
+               *scope = TASK_RUSECPU_FLAGS_DEADLINE;
+               *deadlinep = task->rusage_cpu_deadline;
+       } else {
+               *scope = 0;
+       }
+
+       return 0;
+}
+
+/*
+ * Suspend the CPU usage monitor for the task.  Return value indicates
+ * if the mechanism was actually enabled.
+ */
+int
+task_suspend_cpumon(task_t task)
+{
+       thread_t thread;
+
+       task_lock_assert_owned(task);
+
+       if ((task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_PERTHR_LIMIT) == 0) {
+               return KERN_INVALID_ARGUMENT;
+       }
+
+#if CONFIG_TELEMETRY
+       /*
+        * Disable task-wide telemetry if it was ever enabled by the CPU usage
+        * monitor's warning zone.
+        */
+       telemetry_task_ctl_locked(task, TF_CPUMON_WARNING, 0);
+#endif
+
+       /*
+        * Suspend monitoring for the task, and propagate that change to each thread.
+        */
+       task->rusage_cpu_flags &= ~(TASK_RUSECPU_FLAGS_PERTHR_LIMIT | TASK_RUSECPU_FLAGS_FATAL_CPUMON);
+       queue_iterate(&task->threads, thread, thread_t, task_threads) {
+               act_set_astledger(thread);
+       }
+
+       return KERN_SUCCESS;
+}
+
+/*
+ * Remove all traces of the CPU monitor.
+ */
+int
+task_disable_cpumon(task_t task)
+{
+       int kret;
+
+       task_lock_assert_owned(task);
+
+       kret = task_suspend_cpumon(task);
+       if (kret) {
+               return kret;
+       }
+
+       /* Once we clear these values, the monitor can't be resumed */
+       task->rusage_cpu_perthr_percentage = 0;
+       task->rusage_cpu_perthr_interval = 0;
+
+       return KERN_SUCCESS;
+}
+
+
+static int
+task_enable_cpumon_locked(task_t task)
+{
+       thread_t thread;
+       task_lock_assert_owned(task);
+
+       if (task->rusage_cpu_perthr_percentage == 0 ||
+           task->rusage_cpu_perthr_interval == 0) {
+               return KERN_INVALID_ARGUMENT;
+       }
+
+       task->rusage_cpu_flags |= TASK_RUSECPU_FLAGS_PERTHR_LIMIT;
+       queue_iterate(&task->threads, thread, thread_t, task_threads) {
+               act_set_astledger(thread);
+       }
+
+       return KERN_SUCCESS;
+}
+
+int
+task_resume_cpumon(task_t task)
+{
+       kern_return_t kret;
+
+       if (!task) {
+               return EINVAL;
+       }
+
+       task_lock(task);
+       kret = task_enable_cpumon_locked(task);
+       task_unlock(task);
+
+       return kret;
+}
+
+
+/* duplicate values from bsd/sys/process_policy.h */
+#define PROC_POLICY_CPUMON_DISABLE      0xFF
+#define PROC_POLICY_CPUMON_DEFAULTS     0xFE
+
+static int
+task_set_cpuusage(task_t task, uint8_t percentage, uint64_t interval, uint64_t deadline, int scope, int cpumon_entitled)
+{
+       uint64_t abstime = 0;
+       uint64_t limittime = 0;
+
+       lck_mtx_assert(&task->lock, LCK_MTX_ASSERT_OWNED);
+
+       /* By default, refill once per second */
+       if (interval == 0) {
+               interval = NSEC_PER_SEC;
+       }
+
+       if (percentage != 0) {
+               if (scope == TASK_RUSECPU_FLAGS_PERTHR_LIMIT) {
+                       boolean_t warn = FALSE;
+
+                       /*
+                        * A per-thread CPU limit on a task generates an exception
+                        * (LEDGER_ACTION_EXCEPTION) if any one thread in the task
+                        * exceeds the limit.
+                        */
+
+                       if (percentage == PROC_POLICY_CPUMON_DISABLE) {
+                               if (cpumon_entitled) {
+                                       /* 25095698 - task_disable_cpumon() should be reliable */
+                                       task_disable_cpumon(task);
+                                       return 0;
+                               }
+
+                               /*
+                                * This task wishes to disable the CPU usage monitor, but it's
+                                * missing the required entitlement:
+                                *     com.apple.private.kernel.override-cpumon
+                                *
+                                * Instead, treat this as a request to reset its params
+                                * back to the defaults.
+                                */
+                               warn = TRUE;
+                               percentage = PROC_POLICY_CPUMON_DEFAULTS;
+                       }
+
+                       if (percentage == PROC_POLICY_CPUMON_DEFAULTS) {
+                               percentage = proc_max_cpumon_percentage;
+                               interval   = proc_max_cpumon_interval;
+                       }
+
+                       if (percentage > 100) {
+                               percentage = 100;
+                       }
+
+                       /*
+                        * Passing in an interval of -1 means either:
+                        * - Leave the interval as-is, if there's already a per-thread
+                        *   limit configured
+                        * - Use the system default.
+                        */
+                       if (interval == -1ULL) {
+                               if (task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_PERTHR_LIMIT) {
+                                       interval = task->rusage_cpu_perthr_interval;
+                               } else {
+                                       interval = proc_max_cpumon_interval;
+                               }
+                       }
+
+                       /*
+                        * Enforce global caps on CPU usage monitor here if the process is not
+                        * entitled to escape the global caps.
+                        */
+                       if ((percentage > proc_max_cpumon_percentage) && (cpumon_entitled == 0)) {
+                               warn = TRUE;
+                               percentage = proc_max_cpumon_percentage;
+                       }
+
+                       if ((interval > proc_max_cpumon_interval) && (cpumon_entitled == 0)) {
+                               warn = TRUE;
+                               interval = proc_max_cpumon_interval;
+                       }
+
+                       if (warn) {
+                               int       pid = 0;
+                               const char *procname = "unknown";
+
+#ifdef MACH_BSD
+                               pid = proc_selfpid();
+                               if (current_task()->bsd_info != NULL) {
+                                       procname = proc_name_address(current_task()->bsd_info);
+                               }
+#endif
+
+                               printf("process %s[%d] denied attempt to escape CPU monitor"
+                                   " (missing required entitlement).\n", procname, pid);
+                       }
+
+                       /* configure the limit values */
+                       task->rusage_cpu_perthr_percentage = percentage;
+                       task->rusage_cpu_perthr_interval = interval;
+
+                       /* and enable the CPU monitor */
+                       (void)task_enable_cpumon_locked(task);
+               } else if (scope == TASK_RUSECPU_FLAGS_PROC_LIMIT) {
+                       /*
+                        * Currently, a proc-wide CPU limit always blocks if the limit is
+                        * exceeded (LEDGER_ACTION_BLOCK).
+                        */
+                       task->rusage_cpu_flags |= TASK_RUSECPU_FLAGS_PROC_LIMIT;
+                       task->rusage_cpu_percentage = percentage;
+                       task->rusage_cpu_interval = interval;
+
+                       limittime = (interval * percentage) / 100;
+                       nanoseconds_to_absolutetime(limittime, &abstime);
+
+                       ledger_set_limit(task->ledger, task_ledgers.cpu_time, abstime, 0);
+                       ledger_set_period(task->ledger, task_ledgers.cpu_time, interval);
+                       ledger_set_action(task->ledger, task_ledgers.cpu_time, LEDGER_ACTION_BLOCK);
+               }
+       }
+
+       if (deadline != 0) {
+               assert(scope == TASK_RUSECPU_FLAGS_DEADLINE);
+
+               /* if already in use, cancel and wait for it to cleanout */
+               if (task->rusage_cpu_callt != NULL) {
+                       task_unlock(task);
+                       thread_call_cancel_wait(task->rusage_cpu_callt);
+                       task_lock(task);
+               }
+               if (task->rusage_cpu_callt == NULL) {
+                       task->rusage_cpu_callt = thread_call_allocate_with_priority(task_action_cpuusage, (thread_call_param_t)task, THREAD_CALL_PRIORITY_KERNEL);
+               }
+               /* setup callout */
+               if (task->rusage_cpu_callt != 0) {
+                       uint64_t save_abstime = 0;
+
+                       task->rusage_cpu_flags |= TASK_RUSECPU_FLAGS_DEADLINE;
+                       task->rusage_cpu_deadline = deadline;
+
+                       nanoseconds_to_absolutetime(deadline, &abstime);
+                       save_abstime = abstime;
+                       clock_absolutetime_interval_to_deadline(save_abstime, &abstime);
+                       thread_call_enter_delayed(task->rusage_cpu_callt, abstime);
+               }
+       }
+
+       return 0;
+}
+
+int
+task_clear_cpuusage(task_t task, int cpumon_entitled)
+{
+       int retval = 0;
+
+       task_lock(task);
+       retval = task_clear_cpuusage_locked(task, cpumon_entitled);
+       task_unlock(task);
+
+       return retval;
+}
+
+static int
+task_clear_cpuusage_locked(task_t task, int cpumon_entitled)
+{
+       thread_call_t savecallt;
+
+       /* cancel percentage handling if set */
+       if (task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_PROC_LIMIT) {
+               task->rusage_cpu_flags &= ~TASK_RUSECPU_FLAGS_PROC_LIMIT;
+               ledger_set_limit(task->ledger, task_ledgers.cpu_time, LEDGER_LIMIT_INFINITY, 0);
+               task->rusage_cpu_percentage = 0;
+               task->rusage_cpu_interval = 0;
+       }
+
+       /*
+        * Disable the CPU usage monitor.
+        */
+       if (cpumon_entitled) {
+               task_disable_cpumon(task);
+       }
+
+       /* cancel deadline handling if set */
+       if (task->rusage_cpu_flags & TASK_RUSECPU_FLAGS_DEADLINE) {
+               task->rusage_cpu_flags &= ~TASK_RUSECPU_FLAGS_DEADLINE;
+               if (task->rusage_cpu_callt != 0) {
+                       savecallt = task->rusage_cpu_callt;
+                       task->rusage_cpu_callt = NULL;
+                       task->rusage_cpu_deadline = 0;
+                       task_unlock(task);
+                       thread_call_cancel_wait(savecallt);
+                       thread_call_free(savecallt);
+                       task_lock(task);
+               }
+       }
+       return 0;
+}
+
+/* called by ledger unit to enforce action due to resource usage criteria being met */
+static void
+task_action_cpuusage(thread_call_param_t param0, __unused thread_call_param_t param1)
+{
+       task_t task = (task_t)param0;
+       (void)task_apply_resource_actions(task, TASK_POLICY_CPU_RESOURCE_USAGE);
+       return;
+}
+
+
+/*
+ * Routines for taskwatch and pidbind
+ */
+
+#if CONFIG_TASKWATCH
+
+LCK_MTX_DECLARE_ATTR(task_watch_mtx, &task_lck_grp, &task_lck_attr);
+
+static void
+task_watch_lock(void)
+{
+       lck_mtx_lock(&task_watch_mtx);
+}
+
+static void
+task_watch_unlock(void)
+{
+       lck_mtx_unlock(&task_watch_mtx);
+}
+
+static void
+add_taskwatch_locked(task_t task, task_watch_t * twp)
+{
+       queue_enter(&task->task_watchers, twp, task_watch_t *, tw_links);
+       task->num_taskwatchers++;
+}
+
+static void
+remove_taskwatch_locked(task_t task, task_watch_t * twp)
+{
+       queue_remove(&task->task_watchers, twp, task_watch_t *, tw_links);
+       task->num_taskwatchers--;
+}
+
+
+int
+proc_lf_pidbind(task_t curtask, uint64_t tid, task_t target_task, int bind)
+{
+       thread_t target_thread = NULL;
+       int ret = 0, setbg = 0;
+       task_watch_t *twp = NULL;
+       task_t task = TASK_NULL;
+
+       target_thread = task_findtid(curtask, tid);
+       if (target_thread == NULL) {
+               return ESRCH;
+       }
+       /* holds thread reference */
+
+       if (bind != 0) {
+               /* task is still active ? */
+               task_lock(target_task);
+               if (target_task->active == 0) {
+                       task_unlock(target_task);
+                       ret = ESRCH;
+                       goto out;
+               }
+               task_unlock(target_task);
+
+               twp = (task_watch_t *)kalloc(sizeof(task_watch_t));
+               if (twp == NULL) {
+                       task_watch_unlock();
+                       ret = ENOMEM;
+                       goto out;
+               }
+
+               bzero(twp, sizeof(task_watch_t));
+
+               task_watch_lock();
+
+               if (target_thread->taskwatch != NULL) {
+                       /* already bound to another task */
+                       task_watch_unlock();
+
+                       kfree(twp, sizeof(task_watch_t));
+                       ret = EBUSY;
+                       goto out;
+               }
+
+               task_reference(target_task);
+
+               setbg = proc_get_effective_task_policy(target_task, TASK_POLICY_WATCHERS_BG);
+
+               twp->tw_task = target_task;             /* holds the task reference */
+               twp->tw_thread = target_thread;         /* holds the thread reference */
+               twp->tw_state = setbg;
+               twp->tw_importance = target_thread->importance;
+
+               add_taskwatch_locked(target_task, twp);
+
+               target_thread->taskwatch = twp;
+
+               task_watch_unlock();
+
+               if (setbg) {
+                       set_thread_appbg(target_thread, setbg, INT_MIN);
+               }
+
+               /* retain the thread reference as it is in twp */
+               target_thread = NULL;
+       } else {
+               /* unbind */
+               task_watch_lock();
+               if ((twp = target_thread->taskwatch) != NULL) {
+                       task = twp->tw_task;
+                       target_thread->taskwatch = NULL;
+                       remove_taskwatch_locked(task, twp);
+
+                       task_watch_unlock();
+
+                       task_deallocate(task);                  /* drop task ref in twp */
+                       set_thread_appbg(target_thread, 0, twp->tw_importance);
+                       thread_deallocate(target_thread);       /* drop thread ref in twp */
+                       kfree(twp, sizeof(task_watch_t));
+               } else {
+                       task_watch_unlock();
+                       ret = 0;                /* return success if it not alredy bound */
+                       goto out;
+               }
+       }
+out:
+       thread_deallocate(target_thread);       /* drop thread ref acquired in this routine */
+       return ret;
+}
+
+static void
+set_thread_appbg(thread_t thread, int setbg, __unused int importance)
+{
+       int enable = (setbg ? TASK_POLICY_ENABLE : TASK_POLICY_DISABLE);
+
+       proc_set_thread_policy(thread, TASK_POLICY_ATTRIBUTE, TASK_POLICY_PIDBIND_BG, enable);
+}
+
+static void
+apply_appstate_watchers(task_t task)
+{
+       int numwatchers = 0, i, j, setbg;
+       thread_watchlist_t * threadlist;
+       task_watch_t * twp;
+
+retry:
+       /* if no watchers on the list return */
+       if ((numwatchers = task->num_taskwatchers) == 0) {
+               return;
+       }
+
+       threadlist = kheap_alloc(KHEAP_TEMP,
+           numwatchers * sizeof(thread_watchlist_t), Z_WAITOK | Z_ZERO);
+       if (threadlist == NULL) {
+               return;
+       }
+
+       task_watch_lock();
+       /*serialize application of app state changes */
+
+       if (task->watchapplying != 0) {
+               lck_mtx_sleep(&task_watch_mtx, LCK_SLEEP_DEFAULT, &task->watchapplying, THREAD_UNINT);
+               task_watch_unlock();
+               kheap_free(KHEAP_TEMP, threadlist, numwatchers * sizeof(thread_watchlist_t));
+               goto retry;
+       }
+
+       if (numwatchers != task->num_taskwatchers) {
+               task_watch_unlock();
+               kheap_free(KHEAP_TEMP, threadlist, numwatchers * sizeof(thread_watchlist_t));
+               goto retry;
+       }
+
+       setbg = proc_get_effective_task_policy(task, TASK_POLICY_WATCHERS_BG);
+
+       task->watchapplying = 1;
+       i = 0;
+       queue_iterate(&task->task_watchers, twp, task_watch_t *, tw_links) {
+               threadlist[i].thread = twp->tw_thread;
+               thread_reference(threadlist[i].thread);
+               if (setbg != 0) {
+                       twp->tw_importance = twp->tw_thread->importance;
+                       threadlist[i].importance = INT_MIN;
+               } else {
+                       threadlist[i].importance = twp->tw_importance;
+               }
+               i++;
+               if (i > numwatchers) {
+                       break;
+               }
+       }
+
+       task_watch_unlock();
+
+       for (j = 0; j < i; j++) {
+               set_thread_appbg(threadlist[j].thread, setbg, threadlist[j].importance);
+               thread_deallocate(threadlist[j].thread);
+       }
+       kheap_free(KHEAP_TEMP, threadlist, numwatchers * sizeof(thread_watchlist_t));
+
+
+       task_watch_lock();
+       task->watchapplying = 0;
+       thread_wakeup_one(&task->watchapplying);
+       task_watch_unlock();
+}
+
+void
+thead_remove_taskwatch(thread_t thread)
+{
+       task_watch_t * twp;
+       int importance = 0;
+
+       task_watch_lock();
+       if ((twp = thread->taskwatch) != NULL) {
+               thread->taskwatch = NULL;
+               remove_taskwatch_locked(twp->tw_task, twp);
+       }
+       task_watch_unlock();
+       if (twp != NULL) {
+               thread_deallocate(twp->tw_thread);
+               task_deallocate(twp->tw_task);
+               importance = twp->tw_importance;
+               kfree(twp, sizeof(task_watch_t));
+               /* remove the thread and networkbg */
+               set_thread_appbg(thread, 0, importance);
+       }
+}
+
+void
+task_removewatchers(task_t task)
+{
+       queue_head_t queue;
+       task_watch_t *twp;
+
+       task_watch_lock();
+       queue_new_head(&task->task_watchers, &queue, task_watch_t *, tw_links);
+       queue_init(&task->task_watchers);
+
+       queue_iterate(&queue, twp, task_watch_t *, tw_links) {
+               /*
+                * Since the linkage is removed and thead state cleanup is already set up,
+                * remove the refernce from the thread.
+                */
+               twp->tw_thread->taskwatch = NULL;       /* removed linkage, clear thread holding ref */
+       }
+
+       task->num_taskwatchers = 0;
+       task_watch_unlock();
+
+       while (!queue_empty(&queue)) {
+               queue_remove_first(&queue, twp, task_watch_t *, tw_links);
+               /* remove thread and network bg */
+               set_thread_appbg(twp->tw_thread, 0, twp->tw_importance);
+               thread_deallocate(twp->tw_thread);
+               task_deallocate(twp->tw_task);
+               kfree(twp, sizeof(task_watch_t));
+       }
+}
+#endif /* CONFIG_TASKWATCH */
+
+/*
+ * Routines for importance donation/inheritance/boosting
+ */
+
+static void
+task_importance_update_live_donor(task_t target_task)
+{
+#if IMPORTANCE_INHERITANCE
+
+       ipc_importance_task_t task_imp;
+
+       task_imp = ipc_importance_for_task(target_task, FALSE);
+       if (IIT_NULL != task_imp) {
+               ipc_importance_task_update_live_donor(task_imp);
+               ipc_importance_task_release(task_imp);
+       }
+#endif /* IMPORTANCE_INHERITANCE */
+}
+
+void
+task_importance_mark_donor(task_t task, boolean_t donating)
+{
+#if IMPORTANCE_INHERITANCE
+       ipc_importance_task_t task_imp;
+
+       task_imp = ipc_importance_for_task(task, FALSE);
+       if (IIT_NULL != task_imp) {
+               ipc_importance_task_mark_donor(task_imp, donating);
+               ipc_importance_task_release(task_imp);
+       }
+#endif /* IMPORTANCE_INHERITANCE */
+}
+
+void
+task_importance_mark_live_donor(task_t task, boolean_t live_donating)
+{
+#if IMPORTANCE_INHERITANCE
+       ipc_importance_task_t task_imp;
+
+       task_imp = ipc_importance_for_task(task, FALSE);
+       if (IIT_NULL != task_imp) {
+               ipc_importance_task_mark_live_donor(task_imp, live_donating);
+               ipc_importance_task_release(task_imp);
+       }
+#endif /* IMPORTANCE_INHERITANCE */
+}
+
+void
+task_importance_mark_receiver(task_t task, boolean_t receiving)
+{
+#if IMPORTANCE_INHERITANCE
+       ipc_importance_task_t task_imp;
+
+       task_imp = ipc_importance_for_task(task, FALSE);
+       if (IIT_NULL != task_imp) {
+               ipc_importance_task_mark_receiver(task_imp, receiving);
+               ipc_importance_task_release(task_imp);
+       }
+#endif /* IMPORTANCE_INHERITANCE */
+}
+
+void
+task_importance_mark_denap_receiver(task_t task, boolean_t denap)
+{
+#if IMPORTANCE_INHERITANCE
+       ipc_importance_task_t task_imp;
+
+       task_imp = ipc_importance_for_task(task, FALSE);
+       if (IIT_NULL != task_imp) {
+               ipc_importance_task_mark_denap_receiver(task_imp, denap);
+               ipc_importance_task_release(task_imp);
+       }
+#endif /* IMPORTANCE_INHERITANCE */
+}
+
+void
+task_importance_reset(__imp_only task_t task)
+{
+#if IMPORTANCE_INHERITANCE
+       ipc_importance_task_t task_imp;
+
+       /* TODO: Lower importance downstream before disconnect */
+       task_imp = task->task_imp_base;
+       ipc_importance_reset(task_imp, FALSE);
+       task_importance_update_live_donor(task);
+#endif /* IMPORTANCE_INHERITANCE */
+}
+
+void
+task_importance_init_from_parent(__imp_only task_t new_task, __imp_only task_t parent_task)
+{
+#if IMPORTANCE_INHERITANCE
+       ipc_importance_task_t new_task_imp = IIT_NULL;
+
+       new_task->task_imp_base = NULL;
+       if (!parent_task) {
+               return;
+       }
+
+       if (task_is_marked_importance_donor(parent_task)) {
+               new_task_imp = ipc_importance_for_task(new_task, FALSE);
+               assert(IIT_NULL != new_task_imp);
+               ipc_importance_task_mark_donor(new_task_imp, TRUE);
+       }
+       if (task_is_marked_live_importance_donor(parent_task)) {
+               if (IIT_NULL == new_task_imp) {
+                       new_task_imp = ipc_importance_for_task(new_task, FALSE);
+               }
+               assert(IIT_NULL != new_task_imp);
+               ipc_importance_task_mark_live_donor(new_task_imp, TRUE);
+       }
+       /* Do not inherit 'receiver' on fork, vfexec or true spawn */
+       if (task_is_exec_copy(new_task) &&
+           task_is_marked_importance_receiver(parent_task)) {
+               if (IIT_NULL == new_task_imp) {
+                       new_task_imp = ipc_importance_for_task(new_task, FALSE);
+               }
+               assert(IIT_NULL != new_task_imp);
+               ipc_importance_task_mark_receiver(new_task_imp, TRUE);
+       }
+       if (task_is_marked_importance_denap_receiver(parent_task)) {
+               if (IIT_NULL == new_task_imp) {
+                       new_task_imp = ipc_importance_for_task(new_task, FALSE);
+               }
+               assert(IIT_NULL != new_task_imp);
+               ipc_importance_task_mark_denap_receiver(new_task_imp, TRUE);
+       }
+       if (IIT_NULL != new_task_imp) {
+               assert(new_task->task_imp_base == new_task_imp);
+               ipc_importance_task_release(new_task_imp);
+       }
+#endif /* IMPORTANCE_INHERITANCE */
+}
+
+#if IMPORTANCE_INHERITANCE
+/*
+ * Sets the task boost bit to the provided value.  Does NOT run the update function.
+ *
+ * Task lock must be held.
+ */
+static void
+task_set_boost_locked(task_t task, boolean_t boost_active)
+{
+#if IMPORTANCE_TRACE
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, (IMPORTANCE_CODE(IMP_BOOST, (boost_active ? IMP_BOOSTED : IMP_UNBOOSTED)) | DBG_FUNC_START),
+           proc_selfpid(), task_pid(task), trequested_0(task), trequested_1(task), 0);
+#endif /* IMPORTANCE_TRACE */
+
+       task->requested_policy.trp_boosted = boost_active;
+
+#if IMPORTANCE_TRACE
+       if (boost_active == TRUE) {
+               DTRACE_BOOST2(boost, task_t, task, int, task_pid(task));
+       } else {
+               DTRACE_BOOST2(unboost, task_t, task, int, task_pid(task));
+       }
+       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, (IMPORTANCE_CODE(IMP_BOOST, (boost_active ? IMP_BOOSTED : IMP_UNBOOSTED)) | DBG_FUNC_END),
+           proc_selfpid(), task_pid(task),
+           trequested_0(task), trequested_1(task), 0);
+#endif /* IMPORTANCE_TRACE */
+}
+
+/*
+ * Sets the task boost bit to the provided value and applies the update.
+ *
+ * Task lock must be held.  Must call update complete after unlocking the task.
+ */
+void
+task_update_boost_locked(task_t task, boolean_t boost_active, task_pend_token_t pend_token)
+{
+       task_set_boost_locked(task, boost_active);
+
+       task_policy_update_locked(task, pend_token);
+}
+
+/*
+ * Check if this task should donate importance.
+ *
+ * May be called without taking the task lock. In that case, donor status can change
+ * so you must check only once for each donation event.
+ */
+boolean_t
+task_is_importance_donor(task_t task)
+{
+       if (task->task_imp_base == IIT_NULL) {
+               return FALSE;
+       }
+       return ipc_importance_task_is_donor(task->task_imp_base);
+}
+
+/*
+ * Query the status of the task's donor mark.
+ */
+boolean_t
+task_is_marked_importance_donor(task_t task)
+{
+       if (task->task_imp_base == IIT_NULL) {
+               return FALSE;
+       }
+       return ipc_importance_task_is_marked_donor(task->task_imp_base);
+}
+
+/*
+ * Query the status of the task's live donor and donor mark.
+ */
+boolean_t
+task_is_marked_live_importance_donor(task_t task)
+{
+       if (task->task_imp_base == IIT_NULL) {
+               return FALSE;
+       }
+       return ipc_importance_task_is_marked_live_donor(task->task_imp_base);
+}
+
+
+/*
+ * This routine may be called without holding task lock
+ * since the value of imp_receiver can never be unset.
+ */
+boolean_t
+task_is_importance_receiver(task_t task)
+{
+       if (task->task_imp_base == IIT_NULL) {
+               return FALSE;
+       }
+       return ipc_importance_task_is_marked_receiver(task->task_imp_base);
+}
+
+/*
+ * Query the task's receiver mark.
+ */
+boolean_t
+task_is_marked_importance_receiver(task_t task)
+{
+       if (task->task_imp_base == IIT_NULL) {
+               return FALSE;
+       }
+       return ipc_importance_task_is_marked_receiver(task->task_imp_base);
+}
+
+/*
+ * This routine may be called without holding task lock
+ * since the value of de-nap receiver can never be unset.
+ */
+boolean_t
+task_is_importance_denap_receiver(task_t task)
+{
+       if (task->task_imp_base == IIT_NULL) {
+               return FALSE;
+       }
+       return ipc_importance_task_is_denap_receiver(task->task_imp_base);
+}
+
+/*
+ * Query the task's de-nap receiver mark.
+ */
+boolean_t
+task_is_marked_importance_denap_receiver(task_t task)
+{
+       if (task->task_imp_base == IIT_NULL) {
+               return FALSE;
+       }
+       return ipc_importance_task_is_marked_denap_receiver(task->task_imp_base);
+}
+
+/*
+ * This routine may be called without holding task lock
+ * since the value of imp_receiver can never be unset.
+ */
+boolean_t
+task_is_importance_receiver_type(task_t task)
+{
+       if (task->task_imp_base == IIT_NULL) {
+               return FALSE;
+       }
+       return task_is_importance_receiver(task) ||
+              task_is_importance_denap_receiver(task);
+}
+
+/*
+ * External importance assertions are managed by the process in userspace
+ * Internal importance assertions are the responsibility of the kernel
+ * Assertions are changed from internal to external via task_importance_externalize_assertion
+ */
+
+int
+task_importance_hold_internal_assertion(task_t target_task, uint32_t count)
+{
+       ipc_importance_task_t task_imp;
+       kern_return_t ret;
+
+       /* may be first time, so allow for possible importance setup */
+       task_imp = ipc_importance_for_task(target_task, FALSE);
+       if (IIT_NULL == task_imp) {
+               return EOVERFLOW;
+       }
+       ret = ipc_importance_task_hold_internal_assertion(task_imp, count);
+       ipc_importance_task_release(task_imp);
+
+       return (KERN_SUCCESS != ret) ? ENOTSUP : 0;
+}
+
+int
+task_importance_hold_file_lock_assertion(task_t target_task, uint32_t count)
+{
+       ipc_importance_task_t task_imp;
+       kern_return_t ret;
+
+       /* may be first time, so allow for possible importance setup */
+       task_imp = ipc_importance_for_task(target_task, FALSE);
+       if (IIT_NULL == task_imp) {
+               return EOVERFLOW;
+       }
+       ret = ipc_importance_task_hold_file_lock_assertion(task_imp, count);
+       ipc_importance_task_release(task_imp);
+
+       return (KERN_SUCCESS != ret) ? ENOTSUP : 0;
+}
+
+int
+task_importance_hold_legacy_external_assertion(task_t target_task, uint32_t count)
+{
+       ipc_importance_task_t task_imp;
+       kern_return_t ret;
+
+       /* must already have set up an importance */
+       task_imp = target_task->task_imp_base;
+       if (IIT_NULL == task_imp) {
+               return EOVERFLOW;
+       }
+       ret = ipc_importance_task_hold_legacy_external_assertion(task_imp, count);
+       return (KERN_SUCCESS != ret) ? ENOTSUP : 0;
+}
+
+int
+task_importance_drop_file_lock_assertion(task_t target_task, uint32_t count)
+{
+       ipc_importance_task_t task_imp;
+       kern_return_t ret;
+
+       /* must already have set up an importance */
+       task_imp = target_task->task_imp_base;
+       if (IIT_NULL == task_imp) {
+               return EOVERFLOW;
+       }
+       ret = ipc_importance_task_drop_file_lock_assertion(target_task->task_imp_base, count);
+       return (KERN_SUCCESS != ret) ? EOVERFLOW : 0;
+}
+
+int
+task_importance_drop_legacy_external_assertion(task_t target_task, uint32_t count)
+{
+       ipc_importance_task_t task_imp;
+       kern_return_t ret;
+
+       /* must already have set up an importance */
+       task_imp = target_task->task_imp_base;
+       if (IIT_NULL == task_imp) {
+               return EOVERFLOW;
+       }
+       ret = ipc_importance_task_drop_legacy_external_assertion(task_imp, count);
+       return (KERN_SUCCESS != ret) ? EOVERFLOW : 0;
+}
+
+static void
+task_add_importance_watchport(task_t task, mach_port_t port, int *boostp)
+{
+       int boost = 0;
+
+       __imptrace_only int released_pid = 0;
+       __imptrace_only int pid = task_pid(task);
+
+       ipc_importance_task_t release_imp_task = IIT_NULL;
+
+       if (IP_VALID(port) != 0) {
+               ipc_importance_task_t new_imp_task = ipc_importance_for_task(task, FALSE);
+
+               ip_lock(port);
+
+               /*
+                * The port must have been marked tempowner already.
+                * This also filters out ports whose receive rights
+                * are already enqueued in a message, as you can't
+                * change the right's destination once it's already
+                * on its way.
+                */
+               if (port->ip_tempowner != 0) {
+                       assert(port->ip_impdonation != 0);
+
+                       boost = port->ip_impcount;
+                       if (IIT_NULL != port->ip_imp_task) {
+                               /*
+                                * if this port is already bound to a task,
+                                * release the task reference and drop any
+                                * watchport-forwarded boosts
+                                */
+                               release_imp_task = port->ip_imp_task;
+                               port->ip_imp_task = IIT_NULL;
+                       }
+
+                       /* mark the port is watching another task (reference held in port->ip_imp_task) */
+                       if (ipc_importance_task_is_marked_receiver(new_imp_task)) {
+                               port->ip_imp_task = new_imp_task;
+                               new_imp_task = IIT_NULL;
+                       }
+               }
+               ip_unlock(port);
+
+               if (IIT_NULL != new_imp_task) {
+                       ipc_importance_task_release(new_imp_task);
+               }
+
+               if (IIT_NULL != release_imp_task) {
+                       if (boost > 0) {
+                               ipc_importance_task_drop_internal_assertion(release_imp_task, boost);
+                       }
+
+                       // released_pid = task_pid(release_imp_task); /* TODO: Need ref-safe way to get pid */
+                       ipc_importance_task_release(release_imp_task);
+               }
+#if IMPORTANCE_TRACE
+               KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, (IMPORTANCE_CODE(IMP_WATCHPORT, 0)) | DBG_FUNC_NONE,
+                   proc_selfpid(), pid, boost, released_pid, 0);
+#endif /* IMPORTANCE_TRACE */
+       }
+
+       *boostp = boost;
+       return;
+}
+
+#endif /* IMPORTANCE_INHERITANCE */
+
+/*
+ * Routines for VM to query task importance
+ */
+
+
+/*
+ * Order to be considered while estimating importance
+ * for low memory notification and purging purgeable memory.
+ */
+#define TASK_IMPORTANCE_FOREGROUND     4
+#define TASK_IMPORTANCE_NOTDARWINBG    1
+
+
+/*
+ * (Un)Mark the task as a privileged listener for memory notifications.
+ * if marked, this task will be among the first to be notified amongst
+ * the bulk of all other tasks when the system enters a pressure level
+ * of interest to this task.
+ */
+int
+task_low_mem_privileged_listener(task_t task, boolean_t new_value, boolean_t *old_value)
+{
+       if (old_value != NULL) {
+               *old_value = (boolean_t)task->low_mem_privileged_listener;
+       } else {
+               task_lock(task);
+               task->low_mem_privileged_listener = (uint32_t)new_value;
+               task_unlock(task);
+       }
+
+       return 0;
+}
+
+/*
+ * Checks if the task is already notified.
+ *
+ * Condition: task lock should be held while calling this function.
+ */
+boolean_t
+task_has_been_notified(task_t task, int pressurelevel)
+{
+       if (task == NULL) {
+               return FALSE;
+       }
+
+       if (pressurelevel == kVMPressureWarning) {
+               return task->low_mem_notified_warn ? TRUE : FALSE;
+       } else if (pressurelevel == kVMPressureCritical) {
+               return task->low_mem_notified_critical ? TRUE : FALSE;
+       } else {
+               return TRUE;
+       }
+}
+
+
+/*
+ * Checks if the task is used for purging.
+ *
+ * Condition: task lock should be held while calling this function.
+ */
+boolean_t
+task_used_for_purging(task_t task, int pressurelevel)
+{
+       if (task == NULL) {
+               return FALSE;
+       }
+
+       if (pressurelevel == kVMPressureWarning) {
+               return task->purged_memory_warn ? TRUE : FALSE;
+       } else if (pressurelevel == kVMPressureCritical) {
+               return task->purged_memory_critical ? TRUE : FALSE;
+       } else {
+               return TRUE;
+       }
+}
+
+
+/*
+ * Mark the task as notified with memory notification.
+ *
+ * Condition: task lock should be held while calling this function.
+ */
+void
+task_mark_has_been_notified(task_t task, int pressurelevel)
+{
+       if (task == NULL) {
+               return;
+       }
+
+       if (pressurelevel == kVMPressureWarning) {
+               task->low_mem_notified_warn = 1;
+       } else if (pressurelevel == kVMPressureCritical) {
+               task->low_mem_notified_critical = 1;
+       }
+}
+
+
+/*
+ * Mark the task as purged.
+ *
+ * Condition: task lock should be held while calling this function.
+ */
+void
+task_mark_used_for_purging(task_t task, int pressurelevel)
+{
+       if (task == NULL) {
+               return;
+       }
+
+       if (pressurelevel == kVMPressureWarning) {
+               task->purged_memory_warn = 1;
+       } else if (pressurelevel == kVMPressureCritical) {
+               task->purged_memory_critical = 1;
+       }
+}
+
+
+/*
+ * Mark the task eligible for low memory notification.
+ *
+ * Condition: task lock should be held while calling this function.
+ */
+void
+task_clear_has_been_notified(task_t task, int pressurelevel)
+{
+       if (task == NULL) {
+               return;
+       }
+
+       if (pressurelevel == kVMPressureWarning) {
+               task->low_mem_notified_warn = 0;
+       } else if (pressurelevel == kVMPressureCritical) {
+               task->low_mem_notified_critical = 0;
+       }
+}
+
+
+/*
+ * Mark the task eligible for purging its purgeable memory.
+ *
+ * Condition: task lock should be held while calling this function.
+ */
+void
+task_clear_used_for_purging(task_t task)
+{
+       if (task == NULL) {
+               return;
+       }
+
+       task->purged_memory_warn = 0;
+       task->purged_memory_critical = 0;
+}
+
+
+/*
+ * Estimate task importance for purging its purgeable memory
+ * and low memory notification.
+ *
+ * Importance is calculated in the following order of criteria:
+ * -Task role : Background vs Foreground
+ * -Boost status: Not boosted vs Boosted
+ * -Darwin BG status.
+ *
+ * Returns: Estimated task importance. Less important task will have lower
+ *          estimated importance.
+ */
+int
+task_importance_estimate(task_t task)
+{
+       int task_importance = 0;
+
+       if (task == NULL) {
+               return 0;
+       }
+
+       if (proc_get_effective_task_policy(task, TASK_POLICY_ROLE) == TASK_FOREGROUND_APPLICATION) {
+               task_importance += TASK_IMPORTANCE_FOREGROUND;
+       }
+
+       if (proc_get_effective_task_policy(task, TASK_POLICY_DARWIN_BG) == 0) {
+               task_importance += TASK_IMPORTANCE_NOTDARWINBG;
+       }
+
+       return task_importance;
+}
+
+boolean_t
+task_has_assertions(task_t task)
+{
+       return task->task_imp_base->iit_assertcnt? TRUE : FALSE;
+}
+
+
+kern_return_t
+send_resource_violation(typeof(send_cpu_usage_violation) sendfunc,
+    task_t violator,
+    struct ledger_entry_info *linfo,
+    resource_notify_flags_t flags)
+{
+#ifndef MACH_BSD
+       return KERN_NOT_SUPPORTED;
+#else
+       kern_return_t   kr = KERN_SUCCESS;
+       proc_t          proc = NULL;
+       posix_path_t    proc_path = "";
+       proc_name_t     procname = "<unknown>";
+       int             pid = -1;
+       clock_sec_t     secs;
+       clock_nsec_t    nsecs;
+       mach_timespec_t timestamp;
+       thread_t        curthread = current_thread();
+       ipc_port_t      dstport = MACH_PORT_NULL;
+
+       if (!violator) {
+               kr = KERN_INVALID_ARGUMENT; goto finish;
+       }
+
+       /* extract violator information */
+       task_lock(violator);
+       if (!(proc = get_bsdtask_info(violator))) {
+               task_unlock(violator);
+               kr = KERN_INVALID_ARGUMENT; goto finish;
+       }
+       (void)mig_strncpy(procname, proc_best_name(proc), sizeof(procname));
+       pid = task_pid(violator);
+       if (flags & kRNFatalLimitFlag) {
+               kr = proc_pidpathinfo_internal(proc, 0, proc_path,
+                   sizeof(proc_path), NULL);
+       }
+       task_unlock(violator);
+       if (kr) {
+               goto finish;
+       }
+
+       /* violation time ~ now */
+       clock_get_calendar_nanotime(&secs, &nsecs);
+       timestamp.tv_sec = (int32_t)secs;
+       timestamp.tv_nsec = (int32_t)nsecs;
+       /* 25567702 tracks widening mach_timespec_t */
+
+       /* send message */
+       kr = host_get_special_port(host_priv_self(), HOST_LOCAL_NODE,
+           HOST_RESOURCE_NOTIFY_PORT, &dstport);
+       if (kr) {
+               goto finish;
+       }
+
+       thread_set_honor_qlimit(curthread);
+       kr = sendfunc(dstport,
+           procname, pid, proc_path, timestamp,
+           linfo->lei_balance, linfo->lei_last_refill,
+           linfo->lei_limit, linfo->lei_refill_period,
+           flags);
+       thread_clear_honor_qlimit(curthread);
+
+       ipc_port_release_send(dstport);
+
+finish:
+       return kr;
+#endif      /* MACH_BSD */
+}
+
+
+/*
+ * Resource violations trace four 64-bit integers.  For K32, two additional
+ * codes are allocated, the first with the low nibble doubled.  So if the K64
+ * code is 0x042, the K32 codes would be 0x044 and 0x45.
+ */
+#ifdef __LP64__
+void
+trace_resource_violation(uint16_t code,
+    struct ledger_entry_info *linfo)
+{
+       KERNEL_DBG_IST_SANE(KDBG_CODE(DBG_MACH, DBG_MACH_RESOURCE, code),
+           linfo->lei_balance, linfo->lei_last_refill,
+           linfo->lei_limit, linfo->lei_refill_period);
+}
+#else /* K32 */
+/* TODO: create/find a trace_two_LLs() for K32 systems */
+#define MASK32 0xffffffff
+void
+trace_resource_violation(uint16_t code,
+    struct ledger_entry_info *linfo)
+{
+       int8_t lownibble = (code & 0x3) * 2;
+       int16_t codeA = (code & 0xffc) | lownibble;
+       int16_t codeB = codeA + 1;
+
+       int32_t balance_high = (linfo->lei_balance >> 32) & MASK32;
+       int32_t balance_low = linfo->lei_balance & MASK32;
+       int32_t last_refill_high = (linfo->lei_last_refill >> 32) & MASK32;
+       int32_t last_refill_low = linfo->lei_last_refill & MASK32;
+
+       int32_t limit_high = (linfo->lei_limit >> 32) & MASK32;
+       int32_t limit_low = linfo->lei_limit & MASK32;
+       int32_t refill_period_high = (linfo->lei_refill_period >> 32) & MASK32;
+       int32_t refill_period_low = linfo->lei_refill_period & MASK32;
+
+       KERNEL_DBG_IST_SANE(KDBG_CODE(DBG_MACH, DBG_MACH_RESOURCE, codeA),
+           balance_high, balance_low,
+           last_refill_high, last_refill_low);
+       KERNEL_DBG_IST_SANE(KDBG_CODE(DBG_MACH, DBG_MACH_RESOURCE, codeB),
+           limit_high, limit_low,
+           refill_period_high, refill_period_low);
 }
+#endif /* K64/K32 */